WO2024002216A1 - 一种服务问题归因方法及装置 - Google Patents

一种服务问题归因方法及装置 Download PDF

Info

Publication number
WO2024002216A1
WO2024002216A1 PCT/CN2023/103688 CN2023103688W WO2024002216A1 WO 2024002216 A1 WO2024002216 A1 WO 2024002216A1 CN 2023103688 W CN2023103688 W CN 2023103688W WO 2024002216 A1 WO2024002216 A1 WO 2024002216A1
Authority
WO
WIPO (PCT)
Prior art keywords
text data
analyzed
service
embedding vector
service problem
Prior art date
Application number
PCT/CN2023/103688
Other languages
English (en)
French (fr)
Inventor
付磊
汪安辉
徐志远
尹帅星
梁玉林
Original Assignee
拉扎斯网络科技(上海)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 拉扎斯网络科技(上海)有限公司 filed Critical 拉扎斯网络科技(上海)有限公司
Publication of WO2024002216A1 publication Critical patent/WO2024002216A1/zh

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/205Parsing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/3331Query processing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/35Clustering; Classification
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/36Creation of semantic tools, e.g. ontology or thesauri
    • G06F16/367Ontology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/02Knowledge representation; Symbolic representation

Definitions

  • This application relates to the field of computer technology, specifically to service problem attribution methods and devices, electronic equipment and computer storage media, service problem attribution model training methods and devices, electronic equipment and computer storage media.
  • the online platform After users use the service, the online platform evaluates the service. In order to quickly analyze the problems existing in the evaluation information provided by users for service information, the online platform uses an analysis model to analyze the evaluation information provided by users and identify the types of problems corresponding to the evaluation information.
  • the above-mentioned method of analyzing the problem type of merchant service information usually has the problem of incorrect recognition results and low recognition accuracy. Therefore, how to improve the accuracy of problem types in analysis and evaluation information is a problem that needs to be solved.
  • Embodiments of the present application provide a service problem attribution method, which includes: obtaining text data to be analyzed for service information; inputting the text data to be analyzed into a service problem attribution model, and obtaining text data to be analyzed for the text data to be analyzed. Service problem classification results; wherein the service problem attribution model is used to obtain the service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed. .
  • inputting the text data to be analyzed into a service problem attribution model to obtain a service problem classification result for the text data to be analyzed includes: inputting the text data to be analyzed into a service problem attribution model.
  • a first embedding vector for the text data to be analyzed is obtained, the first embedding vector includes an embedding vector of the text data to be analyzed and a service knowledge graph embedding vector related to the text data to be analyzed; according to the The first embedding vector obtains the service problem classification result corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining the first embedding vector for the text data to be analyzed includes: inputting the text data to be analyzed into the service In the problem attribution model, obtain the text data embedding vector to be analyzed; query the service knowledge graph embedding vector associated with the text data embedding vector to be analyzed according to the text data embedding vector to be analyzed; according to the text data to be analyzed The embedding vector and the service knowledge graph embedding vector are used to obtain the first embedding vector for the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the first embedding vector includes: obtaining the characteristic information of the text data to be analyzed according to the first embedding vector; For the characteristic information of the text data to be analyzed, query the target service problem classification result that matches the characteristic information of the text data to be analyzed in the service problem classification list of the service problem attribution model as the text data to be analyzed. Corresponding service problem classification results.
  • the method includes: obtaining feature information corresponding to multiple candidate service problem classification results in the service problem classification list of the service problem attribution model; The characteristic information is compared with the characteristic information respectively corresponding to the plurality of candidate service question classification results, and the candidate service question classification result containing the characteristic information of the text data to be analyzed is determined to be the same as the characteristic information of the text data to be analyzed.
  • the matching target service problem classification result is used as the service problem classification result corresponding to the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the first embedding vector includes: obtaining the characteristic information of the text data to be analyzed according to the first embedding vector; For the characteristic information of the text data to be analyzed, query at least one service problem classification result that matches the characteristic information of the text data to be analyzed in the service problem classification list of the service problem attribution model; The service problem classification result is used as the service problem classification result corresponding to the text data to be analyzed.
  • Class results include: obtaining the characteristic information of the text data to be analyzed according to the first embedding vector; and obtaining a large classification list of service problems in the service problem attribution model based on the characteristic information of the text data to be analyzed.
  • the first classification result of the service problem associated with the characteristic information of the text data to be analyzed as the first classification result of the target service problem of the text data to be analyzed; determine the target service problem of the text data to be analyzed Based on the first classification result, query the second classification result of the target service question associated with the characteristic information of the text data to be analyzed in the service question sub-classification list of the first classification result of the target service question; according to The first classification result of the target service problem and the second classification result of the target service problem are used to generate the service problem classification result corresponding to the text data to be analyzed.
  • the service problem attribution model includes a large classification text data feature analysis model, and the large classification text data feature analysis model is used to analyze the feature information of the text data corresponding to the first classification result of the service problem; the basis For the characteristic information of the text data to be analyzed, query the first classification result of the service problem associated with the characteristic information of the text data to be analyzed in the large classification list of service problems in the service problem attribution model, as the The first classification result of the target service question of the text data to be analyzed includes: according to the large classification text data feature analysis model, obtaining the first feature of the text data in the text data to be analyzed for analyzing the first classification result of the service question.
  • query the service question sub-classification list of the first classification result of the target service question and the text to be analyzed includes: obtaining the second characteristic information of the text data used to analyze the second classification result of the service problem in the text data to be analyzed; obtaining the small classification of the service problem.
  • the second candidate feature information of the text data corresponding to the second classification result of each candidate service question in the list combine the second feature information in the text data to be analyzed with the second candidate feature information corresponding to the second classification result of each candidate service question. Compare the two candidate feature information to obtain the second classification result of the target service problem corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model to obtain an embedding vector of the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model , obtain the text embedding vector corresponding to each text unit in the text data to be analyzed, the paragraph embedding vector corresponding to each text unit in the text data to be analyzed, and each text unit in the text data to be analyzed.
  • the corresponding position embedding vector; the text embedding vector corresponding to each text unit in the text data to be analyzed, the paragraph embedding vector corresponding to each text unit in the text data to be analyzed, and each text unit in the The corresponding position embedding vector in the text data to be analyzed is processed to obtain the embedding vector of the text data to be analyzed.
  • querying the service knowledge graph embedding vector associated with the text data embedding vector to be analyzed according to the embedding vector of text data to be analyzed includes: obtaining the embedding vector of text data to be analyzed according to the embedding vector of text data to be analyzed.
  • the text embedding vector corresponding to the unit, the paragraph embedding vector corresponding to each text unit in the service knowledge map, the corresponding position embedding vector of each text unit in the service knowledge map; the service knowledge for the service information The text embedding vector corresponding to each text unit in the graph, the paragraph embedding vector corresponding to each text unit in the service knowledge graph, and the corresponding position embedding vector of each text unit in the service knowledge graph are processed to obtain the The service knowledge graph embedding vector of the service information.
  • the service information is food service information for food services provided by the merchant to the user;
  • the analysis text data for the service information is the user's evaluation information for the food service information;
  • the text data to be analyzed includes: obtaining the text data to be analyzed for food service information sent by the client.
  • the service problem attribution model is specifically used to obtain the service knowledge map for the text data to be analyzed based on the text data to be analyzed, the pinyin information corresponding to the text data to be analyzed, and the service knowledge graph related to the text data to be analyzed. Describe the service problem classification results of the text data to be analyzed.
  • inputting the text data to be analyzed into a service problem attribution model to obtain a service problem classification result for the text data to be analyzed includes: inputting the text data to be analyzed into a service problem attribution model.
  • a second embedding vector for the text data to be analyzed is obtained.
  • the second embedding vector includes the pinyin embedding vector of the text data to be analyzed and the pinyin of the service knowledge graph related to the text data to be analyzed.
  • Embedding vector according to the second embedding vector, obtain the service problem classification result corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining a second embedding vector for the text data to be analyzed includes: inputting the text data to be analyzed into the service In the problem attribution model, the text data embedding vector to be analyzed is obtained, and the text data embedding vector to be analyzed includes the text data to be analyzed.
  • the pinyin embedding vector corresponding to each text unit in the text data query the service knowledge graph embedding vector associated with the embedding vector of the text data to be analyzed, and the service knowledge graph embedding vector includes Pinyin embedding vector corresponding to each text unit in the service knowledge graph; according to the embedding vector of the text data to be analyzed and the embedding vector of the service knowledge graph, obtain a second embedding vector for the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining an embedding vector of the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model. , obtain the text embedding vector corresponding to each text unit in the text data to be analyzed, the pinyin embedding vector corresponding to each text unit, the paragraph embedding vector corresponding to each text unit in the text data to be analyzed, each text Embedding vectors at the corresponding positions of units in the text data to be analyzed; embedding text vectors corresponding to each text unit in the text data to be analyzed, pinyin embedding vectors corresponding to each text unit, and embedding each text unit in the The corresponding paragraph embedding vector in the text data to be analyzed and the corresponding position embedding vector of each text unit in the text data to be analyzed are processed to obtain the embedding vector of the text data to be analyzed.
  • querying the service knowledge graph embedding vector associated with the text data embedding vector to be analyzed according to the embedding vector of text data to be analyzed includes: obtaining the embedding vector of text data to be analyzed according to the embedding vector of text data to be analyzed.
  • the service information embedding vector in the text data to be analyzed includes: obtaining the embedding vector of text data to be analyzed according to the embedding vector of text data to be analyzed.
  • the service information embedding vector in the text data to be analyzed includes: obtaining the embedding vector of text data to be analyzed according to the embedding vector of text data to be analyzed.
  • the service information embedding vector in the text data to be analyzed includes: obtaining the embedding vector of text data to be analyzed according to the embedding vector of text data to be analyzed.
  • the service information embedding vector in the text data to be analyzed includes: obtaining the embedding vector of text data to be analyzed according to the embedding vector of text
  • the text embedding vector corresponding to each text unit in the service knowledge graph for the service information, the pinyin embedding vector corresponding to each text unit, the paragraph embedding vector corresponding to each text unit in the service knowledge graph, each The corresponding position embedding vectors of each text unit in the service knowledge graph are processed to obtain a service knowledge graph embedding vector for the service information.
  • Embodiments of the present application also provide a service problem attribution method, which includes: obtaining text data to be analyzed for service information; inputting the text data to be analyzed into a service problem attribution model, and obtaining text data to be analyzed.
  • the service problem classification result wherein the service problem attribution model is used to obtain the service problem classification result for the text data to be analyzed based on the text data to be analyzed and the pinyin information corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into a service problem attribution model to obtain a service problem classification result for the text data to be analyzed includes: inputting the text data to be analyzed into a service problem attribution model.
  • an embedding vector of the text data to be analyzed is obtained.
  • the embedding vector of the text data to be analyzed includes a text embedding vector corresponding to each text unit in the text data to be analyzed, a pinyin embedding vector corresponding to each text unit, and each The paragraph embedding vector corresponding to the text unit in the text data to be analyzed, and the corresponding position of each text unit in the text data to be analyzed is embedded with a vector; according to the embedding vector of the text data to be analyzed, the text to be analyzed is obtained Service problem classification results corresponding to the data.
  • the input of the text data to be analyzed into the service problem attribution model to obtain the embedding vector of the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model, and The text data to be analyzed is subjected to vectorization processing to obtain the text embedding vector corresponding to each text unit in the text data to be analyzed, and the pinyin embedding vector corresponding to each text unit. Each text unit is in the text data to be analyzed.
  • the corresponding paragraph embedding vector, the corresponding position embedding vector of each text unit in the text data to be analyzed; the text embedding vector corresponding to each text unit in the text data to be analyzed, the pinyin embedding corresponding to each text unit Vector, the corresponding paragraph embedding vector of each text unit in the text data to be analyzed, and the corresponding position embedding vector of each text unit in the text data to be analyzed are processed to obtain the embedding vector of the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the embedding vector of the text data to be analyzed includes: obtaining the service problem classification result of the text data to be analyzed according to the embedding vector of the text data to be analyzed.
  • Feature information according to the feature information of the text data to be analyzed, query the target service problem classification result that matches the feature information of the text data to be analyzed in the service problem classification list of the service problem attribution model, as the The service problem classification results corresponding to the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the embedding vector of the text data to be analyzed includes: obtaining the service problem classification result of the text data to be analyzed according to the embedding vector of the text data to be analyzed.
  • Characteristic information according to the characteristic information of the text data to be analyzed, query the first classification result of the service problem associated with the characteristic information of the text data to be analyzed in the large classification list of service problems in the service problem attribution model , as the first classification result of the target service question of the text data to be analyzed; on the basis of determining the first classification result of the target service question of the text data to be analyzed, based on the service question of the first classification result of the target service question Query the second classification result of the target service question associated with the characteristic information of the text data to be analyzed in the small classification list; generate the second classification result of the target service question according to the first classification result of the target service question and the second classification result of the target service question.
  • the service problem classification results corresponding to the text data to be analyzed.
  • the service problem attribution model is specifically used to obtain the service knowledge map for the text data to be analyzed based on the text data to be analyzed, the pinyin information corresponding to the text data to be analyzed, and the service knowledge graph related to the text data to be analyzed. Describe the service problem classification results of the text data to be analyzed.
  • inputting the text data to be analyzed into a service problem attribution model to obtain a service problem classification result for the text data to be analyzed includes: inputting the text data to be analyzed into a service problem attribution model.
  • a second embedding vector for the text data to be analyzed is obtained, the second embedding vector includes an embedding vector of the text data to be analyzed and a service knowledge graph embedding vector related to the text data to be analyzed; according to the The second embedding vector obtains the service problem classification result corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining a second embedding vector for the text data to be analyzed includes: inputting the text data to be analyzed into the service In the problem attribution model, an embedding vector of the text data to be analyzed is obtained.
  • the embedding vector of the text data to be analyzed includes the pinyin embedding vector corresponding to each text unit in the text data to be analyzed.
  • the service knowledge graph embedding vector includes the pinyin embedding vector corresponding to each text unit in the service knowledge graph; according to the text data to be analyzed The embedding vector and the service knowledge graph embedding vector are used to obtain a second embedding vector for the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the second embedding vector includes: obtaining the characteristic information of the text data to be analyzed according to the second embedding vector; For the characteristic information of the text data to be analyzed, query the first classification result of the service problem associated with the characteristic information of the text data to be analyzed in the large classification list of service problems in the service problem attribution model, as the The first classification result of the target service question of the text data to be analyzed; on the basis of determining the first classification result of the target service question of the text data to be analyzed, in the small classification list of service questions of the first classification result of the target service question Query the second classification result of the target service question associated with the characteristic information of the text data to be analyzed; generate the text data to be analyzed according to the first classification result of the target service question and the second classification result of the target service question Corresponding service problem classification results.
  • Embodiments of the present application also provide a training method for a service problem attribution model, which includes: obtaining a pre-training model for analyzing feature information of text data; based on text data samples for service information and services for the text data samples Sample of the first classification result of the problem, adjust the pre-training model to obtain a large classification text data feature analysis model, the large classification text data feature analysis model is used to analyze the characteristics of the text data corresponding to the first classification result of the service problem information; according to the text data sample for the service information and the second classification result sample of the service problem for the text data sample, the large classification text data feature analysis model is adjusted to obtain a service for analyzing the text data to be analyzed Service problem attribution model for problem classification results.
  • the pre-trained model analyzes the feature information of the text data in the following manner: obtains a text data embedding vector according to the text data, and the text data embedding vector includes the text embedding corresponding to each text unit in the text data. vector, the pinyin embedding vector corresponding to each text unit, the paragraph embedding vector of each text unit in the text data, and the position embedding vector of each text unit in the text data; embedding vectors according to the text data , analyze the characteristic information of the text data.
  • the method further includes: obtaining a service knowledge graph embedding vector associated with the text data according to the text data embedding vector, where the service knowledge graph embedding vector includes each of the service knowledge graph embedding vectors associated with the text data.
  • the pre-training model is adjusted according to the text data sample for service information and the first classification result sample of service issues for the text data sample to obtain a large classification text data feature analysis model, including: Input the text data sample for the service information into the pre-training model, and obtain the first service question first classification result of the text data sample for the service information output by the pre-training model; according to the first The degree of similarity between the first classification result of service issues and the sample of the first classification result of service issues for the text data sample is adjusted to the large classification parameters of service issues on the pre-training model to obtain the large classification text Data feature analysis model.
  • the text data sample for the service information is input into the pre-training model, and the first classification result of the first service question for the text data sample for the service information output by the pre-training model is obtained. , including: inputting the text data sample for service information into the pre-training model, and obtaining the characteristic information of the text data sample; and obtaining the characteristic information output by the pre-training model according to the characteristic information of the text data sample. The first classification result of the first service question of the text data sample of the service information.
  • the large classification text data feature analysis model is adjusted according to the text data sample for the service information and the second classification result sample of the service problem for the text data sample to obtain the analysis model for the analysis.
  • the service problem attribution model of the service problem classification result of text data includes: inputting the text data for the service information into the large classification text data feature analysis model, and obtaining the large classification text data feature analysis model output for the service problem attribution model.
  • the second classification result of the second service question of the text data sample according to the similarity between the second classification result of the second service question and the second classification result sample of the service question, the features of the large classification text data are
  • the analysis model is adjusted to obtain a service problem attribution model used to analyze the service problem classification results for the text data to be analyzed.
  • it also includes: using text data samples for service information and first classification result samples of service issues that do not belong to the text data as a first negative sample pair, adjusting the pre-training model to obtain a sample for analysis A large classification text data feature analysis model corresponding to the feature information of the text data of the first classification result of the service question; using the text data sample for the service information and the second classification result sample of the service question that does not belong to the text data as the second negative For sample pairs, the large classification text data feature analysis model is adjusted to obtain a service problem attribution model used to analyze the service problem classification results of the text data to be analyzed.
  • Embodiments of the present application also provide a service problem attribution device, including: a first obtaining unit for obtaining text data to be analyzed for service information; a second obtaining unit for inputting the text data to be analyzed into the service
  • a service problem classification result for the text data to be analyzed is obtained; wherein the service problem attribution model is used to classify the text data according to the text data to be analyzed and the service knowledge related to the text data to be analyzed.
  • Map to obtain service problem classification results for the text data to be analyzed.
  • Embodiments of the present application also provide a service problem attribution device, including: a third obtaining unit for obtaining text data to be analyzed for service information; a fourth obtaining unit for inputting the text data to be analyzed into the service
  • a service problem classification result for the text data to be analyzed is obtained; wherein the service problem attribution model is used to determine the text data to be analyzed and the pinyin information corresponding to the text data to be analyzed, Obtain service problem classification results for the text data to be analyzed.
  • Embodiments of the present application also provide a training device for a service problem attribution model, including: a pre-training model acquisition unit for obtaining a pre-training model for analyzing feature information of text data; and a large classification text data feature analysis model acquisition unit , used to adjust the pre-training model according to the text data sample for the service information and the first classification result sample of the service problem for the text data sample to obtain a large classification text data feature analysis model, the large classification text
  • the data feature analysis model is used to analyze the feature information of the text data corresponding to the first classification result of the service problem; the service problem attribution model acquisition unit is used to obtain the service problem according to the text data sample for the service information and the service problem for the text data sample.
  • the large classification text data feature analysis model is adjusted to obtain a service problem attribution model used to analyze the service problem classification results for the text data to be analyzed.
  • An embodiment of the present application also provides an electronic device.
  • the electronic device includes a processor and a memory.
  • a computer program is stored in the memory. After the processor runs the computer program, it executes the above method.
  • Embodiments of the present application also provide a computer storage medium, the computer storage medium stores a computer program, and when the computer program is executed, the above method is executed.
  • Embodiments of the present application provide a service problem attribution method, which includes: obtaining text data to be analyzed for service information; inputting the text data to be analyzed into a service problem attribution model, and obtaining text data to be analyzed for the text data to be analyzed. Service problem classification results; wherein the service problem attribution model is used to obtain the service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed. .
  • the service problem attribution model analyzes and obtains service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed.
  • the service problem attribution model provided in the above method uses the text data to be analyzed and the service knowledge graph related to the text data to be analyzed as input information.
  • the service knowledge graph can obtain the attribute information of the service information in the text data to be analyzed, thereby Improve the understanding of language expression logic for analyzing text data. Based on the service knowledge graph, it helps to improve the accuracy of the service problem attribution model in determining the language expression logic of the text data to be analyzed. The accuracy of the service problem attribution model in matching the service problem classification results of the text data to be analyzed is also improved.
  • the service problem attribution model analyzes and obtains service problem classification results for the text data to be analyzed based on the text data to be analyzed and the pinyin information corresponding to the text data to be analyzed.
  • the service problem attribution model provided in the above method, when processing the text data to be analyzed, combines the pinyin information of the text data to be analyzed to improve the accuracy of the language expression logic of the text data to be analyzed. Based on the text data to be analyzed, it helps to improve the accuracy of the language expression logic of the text data to be analyzed, and the service problem attribution model matches the text data to be analyzed. The accuracy of service problem classification results has also been improved.
  • a training method for a service problem attribution model includes: obtaining a pre-trained model for analyzing feature information of text data; Describe the first classification result sample of the service problem of the text data sample, adjust the pre-training model to obtain a large classification text data feature analysis model, and the large classification text data feature analysis model is used to analyze the first classification corresponding to the service problem Characteristic information of the text data of the result; according to the text data sample for the service information and the second classification result sample of the service question for the text data sample, the large classification text data feature analysis model is adjusted to obtain the analysis for Service problem attribution model for service problem classification results of text data to be analyzed.
  • the above method is used to obtain a pre-trained model.
  • text data samples and first classification result samples of service problems for the text data samples to be analyzed are used to adjust the pre-trained model to obtain a large classification text data feature analysis model.
  • the large classification text data feature analysis model is adjusted to obtain the service problem attribution used to analyze the service problem classification result for the text data to be analyzed.
  • the trained service problem attribution model obtains the service problem classification result of the text data to be analyzed, it first obtains the first classification result of the service problem of the text data to be analyzed, and then determines the first classification result of the service problem of the text data to be analyzed. On the basis of the results, among multiple second classification results of service problems corresponding to the first classification result of the target service problem, the second classification result of the service problem of the text data to be analyzed is determined. Therefore, the finally obtained service problem classification result of the text data to be analyzed includes the first classification result of service problem and the second classification result of service problem.
  • Figure 1 is an application scenario diagram of the service problem attribution method provided by the embodiment of the present application.
  • Figure 2 is a schematic diagram of the service problem classification results for text information in the service problem attribution model provided by the embodiment of the present application.
  • Figure 3 is a display page of the merchant physical examination center provided by the embodiment of this application.
  • FIG 4 shows detailed information on food safety incidents in merchant stores provided by the embodiment of this application.
  • Figure 5 is the data information of the food safety incident detailed information in Figure 4.
  • Figure 6 shows the level of food safety incidents that occurred at the merchant in Figure 5 and the corresponding handling methods for the food safety incidents.
  • Figure 7 is a learning method for food safety events developed for merchants provided by the embodiment of this application.
  • Figure 8 is a flow chart of a service problem attribution method provided in the first embodiment of the present application.
  • Figure 9 is a schematic diagram of a service problem attribution device provided in the second embodiment of the present application.
  • Figure 10 is a schematic diagram of another service problem attribution method provided in the third embodiment of the present application.
  • Figure 11 is a schematic diagram of another service problem attribution device provided in the fourth embodiment of the present application.
  • Figure 12 is a schematic diagram of a training method for a service problem attribution model provided in the fifth embodiment of the present application.
  • Figure 13 is a schematic diagram of a training device for a service problem attribution model provided in the sixth embodiment of the present application.
  • Figure 14 is a schematic diagram of an electronic device provided in the seventh embodiment of the present application.
  • the service problem attribution method provided by this application can be applied to food safety problem classification scenarios.
  • an online-to-offline catering service system provides users with meal ordering services.
  • Figure 1 it is an application scenario diagram of the service problem attribution method provided by the embodiment of the present application.
  • Users order meals through the client 101, unsubscribe meals, and user feedback related data for meals, such as evaluation information for unsubscribed meals, food safety insurance claim evaluation information, and evaluation information for consumed meals. wait.
  • the evaluation information includes user complaint information regarding safety issues of some foods.
  • the server 102 receives the evaluation information for the meal sent by the user terminal, and determines based on the evaluation information whether the evaluation information is complaint information for the food safety issues of the meal. If so, determines the information corresponding to the evaluation information.
  • the type of food safety issue, and the evaluation information and the type of food safety issue for the evaluation information are sent to the merchant that provides the meal.
  • the server 102 analyzes the type of food safety problem corresponding to the evaluation information provided by the user, and performs the analysis through the service problem attribution model. Specifically, the obtained text data to be analyzed for the service information is input into the service problem attribution model, and the service problem classification result for the text data to be analyzed outputted by the service problem attribution model is obtained.
  • the user's evaluation information of the meal is input into the service problem attribution model.
  • the service problem attribution model analyzes the evaluation information and determines the service problem classification result corresponding to the evaluation information, which is the food safety problem corresponding to the meal. type.
  • the user's evaluation information for a meal is: "There are mold spots on the fruits in this fruit drink.” If the evaluation information is input into the service problem attribution model, then the service problem attribution model outputs the corresponding value for the meal.
  • the type of food safety problem is visible mold and spoilage. For another example, if the evaluation information is: "There are cockroaches in this meal," and the evaluation information is input into the service problem attribution model, the food safety problem type corresponding to the meal output by the service problem attribution model is cockroaches.
  • the service problem attribution model is used to determine the service problem classification results corresponding to the evaluation information based on the evaluation information provided by the user. Specifically, first determine the large classification result of the service problem corresponding to the evaluation information (also called the first classification result of the service problem), and then analyze the small classification result of the service problem corresponding to the evaluation information based on determining the large classification result of the service problem. Results (also known as service issue secondary classification results). Based on the large classification results of service problems and the small classification results of service problems in the large classification results of service problems, the service problem classification results for the evaluation information are generated. Take the above evaluation information "There are cockroaches in this meal" as an example. The large classification result of the service problem of this evaluation information is foreign objects, and the small classification result of the service problem is cockroaches. Among them, cockroaches are the small classification of pests and mice in the foreign objects. result.
  • the server After the server obtains the food safety issue type corresponding to the user's evaluation information, it sends the food safety incidents in the food provided by the merchant and the service issue classification results corresponding to the food safety incidents to the merchant. Merchants can check their existing food safety incidents through the physical examination center on the merchant side.
  • FIG 3 it is a display page of the merchant-side physical examination center provided by the embodiment of the present application.
  • the physical examination center on the merchant side displays the statistics of safety incidents at the merchant, as well as control and processing records, the merchant's food safety credit score information, and information on the learning area for food safety issues.
  • the page in Figure 3 shows merchants their current food safety incidents and corresponding learning strategies, allowing merchants to clarify their current food safety incidents and further correction plans.
  • the view event button on the page in Figure 3 the page shown in Figure 4 is entered.
  • Figure 4 is detailed information on food safety events in merchant stores provided by the embodiment of the present application.
  • merchants can learn detailed information about specific food safety incidents that occurred within the current month, and merchants can further improve the problems based on this details.
  • Figure 5 is data information of the detailed information of the food safety incident in Figure 4.
  • merchants can understand their food safety incident incidence rate in the current month, as well as the number of food safety incidents.
  • Figure 6 shows the food safety incident level that occurred at the merchant in Figure 5 and the processing method corresponding to the food safety incident level.
  • the server 102 will also provide the merchant's terminal 103 used by the merchant with learning courses for handling the food safety problem type, so that the merchant can handle the food safety problem in a timely manner and prevent continued food safety problems.
  • Figure 7 is a learning method for food safety events developed for merchants provided by the embodiment of this application. Specifically, Figure 7 takes the prevention and control methods of pests, mice and foreign bodies as an example to explain the details of the prevention and control of pests, mice and foreign bodies, the corresponding regulatory requirements and preventive measures.
  • the service problem attribution method provided by the embodiment of the present application can also be applied to the scenario of recommending meals to users.
  • a user logs into an online takeout platform and searches for relevant meal keywords in the search box of the online takeout platform.
  • the server of the online takeout platform uses the corresponding recommendation algorithm based on the user's search keywords to obtain the products corresponding to the search keywords.
  • Meal information Send the obtained meal information to the user terminal, and the user terminal displays the obtained meal information.
  • the server determines the corresponding recommended food based on the search keywords provided by the user
  • the factors referenced by its recommendation algorithm include multiple factors, such as the user's dining habit information, the food feature information of the food searched by the user, The user's geographical location information and the food safety level corresponding to the merchant that provides the meal.
  • the food safety level of the merchant can be obtained by using the service problem attribution model in the service problem attribution method provided by this application.
  • the service problem attribution method provided by the embodiment of the present application can also be applied to the merchant information recommended on the home page of the client.
  • the merchant information displayed on the user's homepage is obtained by the server in advance through a recommendation algorithm.
  • the reference factors of the recommendation algorithm include: the distance information between the geographical location of the merchant and the geographical location of the user, the type of merchants that the user has paid attention to in history, the type of food, and the food safety level of the merchant.
  • the food safety level of the merchant can be obtained by using the service problem attribution model in the service problem attribution method provided by this application.
  • the food safety level of a merchant can be obtained through the following methods:
  • the service problem attribution model is used to analyze the evaluation information provided by the user to determine the food safety issues in the evaluation information and the type of the food safety issues. That is to say, if the evaluation information provided by the user for the meal he or she eats contains food safety issues, the specific service issue classification results corresponding to the food safety issues will be determined based on the service issue attribution model. The food safety level of the merchant corresponding to the food is determined based on the type of food safety problem identified.
  • a service problem attribution model is used to analyze the types of food safety problems contained in user evaluation information.
  • the evaluation information for the meal and the food safety issue type corresponding to the food safety issue in the evaluation information are sent to the merchant that provides the meal to the user.
  • it enhances merchants’ food safety awareness and reminds them of the seriousness of food safety issues.
  • merchants are reminded of specific food safety issues raised by users when eating food.
  • merchants should be provided with methods to deal with such food safety problems to avoid such food safety problems from happening again in the future. Among them, if the merchant has too many types of food safety problems, certain mandatory rectification measures will be provided to the merchant.
  • the online takeout system can determine the food safety level of the merchant. When recommending the corresponding merchant list to the user, it determines whether to recommend the merchant to the user based on the merchant's food safety level.
  • the service problem attribution model mentioned in the above application scenarios analyzes user evaluation information to obtain the service problem classification results corresponding to the user evaluation information.
  • the service problem attribution model is specifically analyzed through the following methods:
  • the service problem attribution model determines the service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph associated with the text data to be analyzed.
  • the service knowledge graph refers to the prior knowledge graph of the text data to be analyzed, which is pre-stored in the service problem attribution model.
  • Service knowledge graphs include at least the following types:
  • the service knowledge graph is an attribution service knowledge graph, which is the knowledge graph of the service information in the text data to be analyzed.
  • the text data to be analyzed is: Stinky tofu smells so bad.
  • the corresponding service knowledge map is: stinky tofu is a kind of food, and the smell attribute of this food is smelly.
  • the analysis result is that the stinky tofu has a deterioration problem.
  • the service problem attribution model analyzes the text data to be analyzed "Stinky tofu smells so bad", it will also combine it with the service knowledge graph corresponding to the text data to be analyzed "Stinky tofu is a kind of food. "The smell attribute of food is smelly”, analyze the service problem classification results corresponding to the text data to be analyzed.
  • the information "Stinky tofu smells so bad” and "Stinky tofu is a kind of food, and the smell attribute of this food is stinky" can be determined that the problem fed back in the text data to be analyzed does not belong to the service problem attribution Any attribution type in the model's attribution criteria indicates that the problem reported by the text data to be analyzed is not a service problem.
  • the service problem attribution model vectorizes the input information, and the output information is also represented by vectors. For example, a vector of 1 indicates that the service information of the text data to be analyzed corresponds to at least one attribution type in the attribution standard. A vector of 0 indicates that the service information of the text data to be analyzed does not correspond to any attribution type in the attribution standard.
  • the text data to be analyzed is: This snail rice noodle smells so bad.
  • Traditional analysis methods will determine that the snail powder has a deterioration problem.
  • the service problem attribution model is combined with the service knowledge graph when analyzing the text data to be analyzed.
  • the service knowledge graph corresponding to the text data to be analyzed is: Snail rice noodle is a commodity, and the smell attribute of this commodity is smelly.
  • the service problem attribution model combines the analysis of the text data to be analyzed with its corresponding service knowledge graph, and concludes that the smell of snail noodles is a product attribute of snail noodles itself, indicating that the problems reported in the text data to be analyzed do not belong to the service Any of the attribution types within the attribution criteria of the problem attribution model.
  • the service knowledge map is an ingredient map. For example, if the text data to be analyzed is "Why is there black pepper in the black pepper beef fillet?", the corresponding service knowledge map is "The ingredient list of the black pepper beef fillet includes: black pepper.” Wagyu Fillet”.
  • the service problem attribution model will combine it with the corresponding service knowledge graph "Ingredients list of black pepper beef tenderloin” Included: Black Pepper and Beef Tenderloin.” It was determined from this that the black pepper in the black pepper beef tenderloin is one of the main ingredients of this dish. Therefore, it means that the problem reported by the text data to be analyzed does not belong to any of the attribution types in the attribution standards in the service problem attribution model, and there is no service problem in it.
  • the service knowledge map is a recipe. For example, if the text data to be analyzed is "What kind of fish is pickled fish?", the corresponding service knowledge map is "Sauerkraut fish is a recipe, and the main ingredients are pickled cabbage and fish.”
  • the analysis result is that the pickled fish problem is a food spoilage problem.
  • the service problem attribution model will combine it with its corresponding service knowledge graph "Sauerkraut fish is a recipe, and the main ingredients are pickled cabbage and fish.” ". It was determined from this that pickled fish is a dish made from ordinary fish and pickled cabbage, not the name of the fish species. Therefore, it means that the problem reported in the text data to be analyzed does not belong to any of the attribution types in the attribution criteria in the service problem attribution model, and there is no service problem.
  • the service knowledge graph is a common sense graph. For example, the text data to be analyzed is "rats are doing it", and the corresponding service knowledge map is "haozi is doing it".
  • the service problem attribution model will combine it with its corresponding service knowledge graph "Hao Ziwei" when analyzing the text data to be analyzed "Mouse is for it". It is determined from this that the text data to be analyzed is a reminder from the user to the merchant that compared with the meal service provided by the merchant before to the user, the service this time is relatively poor, but it does not fall into any of the attribution criteria in the service problem attribution model. This is a type of attribution, and we hope that merchants will take good care of themselves and improve their current service quality.
  • the above are several common types of service knowledge graphs.
  • the service problem attribution model analyzes the language expression logic of the text data to be analyzed, combined with the service knowledge graph corresponding to the text data to be analyzed, it can improve the language expression logic of the text data to be analyzed. Accuracy. Furthermore, after obtaining the language expression logic of the text data to be analyzed based on the text data to be analyzed and the service knowledge graph corresponding to the text data to be analyzed, the accuracy of determining the service classification results corresponding to the text data to be analyzed is also improved.
  • the service problem attribution model uses the text data to be analyzed and the service knowledge graph associated with the text data to be analyzed.
  • the process of determining the service problem attribution results corresponding to the text data to be analyzed is as follows:
  • the text data embedding vector to be analyzed includes the text embedding vector corresponding to each text unit in the text data to be analyzed.
  • Each text unit is located at The paragraph embedding vector in the text data to be analyzed, and the position embedding vector of each text unit in the text data to be analyzed.
  • each text unit can be each word or each word.
  • E represents the text embedding vector corresponding to the word "this”
  • EA represents the paragraph embedding vector of the word "this” in the text data to be analyzed
  • E1 represents the paragraph embedding vector of the word "this” in the text data to be analyzed. Analyze positional embedding vectors in text data.
  • the service knowledge graph embedding vector includes the text embedding vector corresponding to each text unit in the service knowledge graph.
  • Each text The paragraph embedding vector of the unit in the service knowledge graph, and the position embedding vector of each text unit in the service knowledge graph.
  • the text data to be analyzed embedding vector and the service knowledge graph embedding vector associated with the text data to be analyzed are used to generate a first embedding vector for the text data to be analyzed.
  • the service problem attribution model analyzes the first embedding vector of the text data to be analyzed, determines the characteristic information of the text data to be analyzed, and uses it to analyze the service problem classification results. Based on the characteristic information of the text data to be analyzed, determine the service corresponding to the text data to be analyzed. Problem classification results.
  • the service problem analysis model determines the service problem classification results corresponding to the text data to be analyzed based on the characteristic information of the text data to be analyzed.
  • the classification results of service problems corresponding to the text data to be analyzed are determined based on the characteristic information of the text data to be analyzed, such as the classification results of foreign matter, spoilage, food-borne diseases, expiration, undercooked food, etc. shown in Figure 2.
  • the process of determining the large classification result of the target service problem (also called the first classification result of the target service problem) is also called the fixed sharing layer.
  • the target classification layer corresponding to the characteristic information of the text data to be analyzed, that is, determine the small classification result of the target service problem of the text data to be analyzed (also known as the second classification result of the target service problem) .
  • the determined small classification result of the target service problem is the classification result of the target service problem of the text data to be analyzed.
  • the service problem attribution model determines the service problem classification results for the text data to be analyzed based on the text data to be analyzed and the pinyin information of the text data to be analyzed.
  • the text data to be analyzed is input into the service problem attribution model to obtain the text data embedding vector to be analyzed.
  • the text data embedding vector to be analyzed includes the text embedding vector corresponding to each text unit in the text data to be analyzed.
  • Each text unit The corresponding pinyin embedding vector, the paragraph embedding of each text unit in the text data to be analyzed Vector, the position of each text unit in the text data to be analyzed is embedded in the vector.
  • the embedding vector of the text data to be analyzed includes the pinyin embedding vector corresponding to each text unit. Therefore, the service problem attribution model converts the text data to be analyzed into an embedding vector, which can improve the phenomenon of typos and thereby improve the language expression of the text data to be analyzed. logical accuracy.
  • the service problem attribution model determines the service problem classification for the text data to be analyzed based on the text data to be analyzed, the pinyin information of the text data to be analyzed, and the service knowledge graph associated with the text data to be analyzed. result.
  • the embedding vector of the text data to be analyzed includes the text embedding vector corresponding to each text unit in the text data to be analyzed, and the pinyin corresponding to each text unit. Embedding vector, the paragraph embedding vector of each text unit in the text data to be analyzed, and the position embedding vector of each text unit in the text data to be analyzed.
  • E rice is the text embedding vector corresponding to the word rice
  • E A is the paragraph embedding vector of the word rice in the text data to be analyzed in "There is a cockroach in this meal”
  • E3 is the word rice in "this "There are cockroaches in the food” is the position embedding vector in the text data to be analyzed
  • E fan is the pinyin embedding vector corresponding to the word "rice”.
  • the service knowledge graph embedding vector associated with the text data to be analyzed is determined.
  • the service knowledge graph embedding vector includes: the text embedding vector corresponding to each text unit in the service knowledge graph, and the text embedding vector corresponding to each text unit. Pinyin embedding vector, paragraph embedding vector of each text unit in the service knowledge graph, position embedding vector of each text unit in the service knowledge graph.
  • the service problem attribution model analyzes the text data to be analyzed, on the basis of determining the text embedding vector, paragraph embedding vector and position embedding vector of the text data to be analyzed, it combines the pinyin embedding vector to reduce the frequency of typos.
  • the service knowledge graph associated with the text data to be analyzed the language expression logic of the text data to be analyzed is analyzed. Adding pinyin embedding vectors and service knowledge graphs can improve the accuracy of analyzing the language expression logic of the text data to be analyzed. Therefore, the service problem attribution model can extract the text data to be analyzed for analysis based on the language expression logic of the text data to be analyzed. Characteristic data of service problem classification results.
  • the service problem classification result corresponding to the characteristic data is determined, and then, based on the determined service problem classification result, in the service problem classification result Select the service problem small classification result corresponding to the characteristic information from the service problem small classification list.
  • the service problem classification result corresponding to the text data to be analyzed is determined according to the determined large classification result of the target service problem and the small classification result of the target service problem in the large classification result of the target service problem.
  • the first step is to obtain a pre-trained model for analyzing feature information of text data.
  • the pre-training model can be obtained after training the traditional Bert (Bidirectional Encoder Representations from Transformers, deep bidirectional pre-training encoder) model.
  • the Bert model is a pre-trained language model that can be used in questions and answers systems, sentiment analysis, spam filtering, named entity recognition, document clustering and other tasks.
  • the input object of the Bert model is the original vector of each word in the text to be recognized, including: word vector (Token Embedding, or text vector), paragraph vector (Segment Embedding) and position vector (Position Embedding).
  • word vector Token Embedding, or text vector
  • paragraph vector Segment Embedding
  • position vector Position Embedding
  • the Bert model includes at least three types of inputs and corresponds to three types of outputs. details as follows:
  • Sequence plus annotation classification The output vector corresponding to each word in this type of task is the annotation of the word, which can be understood as classification.
  • the Transformer structure in the Bert model obtains bidirectional information through the multi-head attention mechanism and the masking mechanism to enhance the language expression logic of the text data to be recognized.
  • the attention mechanism allows the neural network to focus on a part of the input.
  • the multi-head attention mechanism linearly combines multiple enhanced semantic vectors for each word, and finally obtains an enhanced semantic vector that is equal to the length of the original word vector. For example, “Is this set meal / very / delicious?" and "Is this set meal / very good / delicious?" In these two sentences, the semantics expressed by the combination of "eat” and "good” are different.
  • the masking mechanism means that when the text to be recognized is input into the model, any words or words in a sentence are randomly masked or replaced, and the model can predict the content of the masked or replaced part of the sentence through contextual understanding. Then, the model determines the attribution classification model corresponding to the text to be recognized based on the emotion reflected in the contextual content of the text to be recognized after identifying the obscured or replaced content.
  • the pre-training model converts the text data to be analyzed into embedding vectors, it adds Pinyin on the basis of the original text embedding vector, paragraph embedding vector and position embedding vector. Embedding vectors to improve the error rate of text recognition. Then the service knowledge graph is embedded in the embedding vector of the text data to be analyzed to improve the accuracy of analyzing the language expression logic of the text data to be analyzed. Therefore, the pre-training model is used to obtain the language expression logic of the text data to be analyzed, and analyze and determine the characteristic information of the text data to be analyzed, which will serve as the basis for subsequent determination of the service problem classification results of the text data to be analyzed.
  • the second step is to adjust the pre-training model according to the text data sample for service information and the large classification result sample of service issues for the text data sample to obtain a large classification text data feature analysis model.
  • the large classification text The data feature analysis model is used to analyze the feature information of text data corresponding to the large classification results of service problems.
  • the training purpose of the above-mentioned pre-training model is to obtain the characteristic information of the text data to be analyzed.
  • text data samples for service information and the text data are used. Samples of large classification results of service problems are used to perform parameter training on the pre-trained model.
  • the large classification text data feature analysis model obtained through training can obtain the large classification results of service issues corresponding to the text data to be analyzed based on the text data to be analyzed.
  • FIG. 2 it is a schematic diagram of the service problem classification results for text information in the service problem attribution model provided by the embodiment of the present application.
  • the results of the broad classification of service problems can be foreign objects, spoilage, food-borne diseases, expired, undercooked food, etc.
  • the third step is to adjust the feature analysis model of the large classification text data according to the text data sample for the service information and the small classification result sample of the service problem for the text data sample, and obtain the feature analysis model for the text data to be analyzed.
  • the large classification text data feature analysis model is used to analyze the large classification results of service problems of text data. On this basis, further parameter adjustments are made to the large classification text data feature analysis model so that it can analyze the small service problems of text data. Classification results.
  • the pre-training model is adjusted using text data samples for service information and large classification result samples of service problems for the text data samples to obtain a large classification text data feature analysis model.
  • the "service problem classification result sample for the text data sample” mentioned here refers to the service problem classification result sample belonging to the text data sample, which is used as a positive sample.
  • the pre-training model can also be trained using text data samples for service information and service problem classification result samples that do not belong to the text data samples as the first negative sample pair.
  • the text data sample is "There are cockroaches in this meal”
  • the service problem classification result sample belonging to the text data sample is "foreign object", which is a positive sample.
  • the service problem classification result sample that does not belong to the text data sample can be "spoilage”, or “foodborne disease”, or “expired”, or "undercooked food”, etc., which are negative samples.
  • Positive samples are the large classification result samples of service problems belonging to text data samples that need to be queried by the large classification text data feature analysis model after training.
  • the negative samples are comparison samples of the positive samples, making the large classification text data feature analysis model unable to identify Large classification results of service problems belonging to text data samples.
  • Positive samples and negative samples are used to train the large classification text data feature analysis model, so that the classification accuracy of the large classification text data sample data feature analysis model for the large classification results of service problems corresponding to the text data is improved.
  • the text data sample for service information and the service problem small classification result sample for the text data sample are used to adjust the large classification text data feature analysis model to obtain a service problem attribution model.
  • the "sample of service problem classification results for the text data sample” mentioned here refers to the service problem classification result belonging to the text data sample, which is used as a positive sample.
  • the service problem small classification result belonging to the text data sample here is at least one result among the multiple candidate service problem small classification results in the service problem large classification result determined in the second step.
  • the text data sample for the service information and the services that do not belong to the text data sample can also be
  • the problem small classification result samples are used as the second negative sample pair to train the large classification text data feature analysis model.
  • the text data sample is "There are cockroaches in this meal”
  • the service problem classification result sample belonging to this text data sample is "foreign object”
  • the service problem classification result sample belonging to this text data sample is "cockroach”.
  • the service problem small classification result sample that does not belong to the text data sample can be "sharp foreign objects", or "rats”, etc., which are negative samples.
  • Positive samples and negative samples are used to train the large classification text data feature analysis model, so that the obtained service problem attribution model can improve the classification accuracy of the service problem classification results corresponding to the text data.
  • Embodiments of the present application provide a service problem attribution method, which includes: obtaining text data to be analyzed for service information; inputting the text data to be analyzed into a service problem attribution model, and obtaining text data to be analyzed for the text data to be analyzed. Service problem classification results; wherein the service problem attribution model is used to obtain the service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed. .
  • the service problem attribution model analyzes and obtains service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed.
  • the service problem attribution model provided in the above method uses the text data to be analyzed and the service knowledge graph related to the text data to be analyzed as input information.
  • the service knowledge graph can obtain the attribute information of the service information in the text data to be analyzed, thereby Improve the understanding of language expression logic for analyzing text data. Based on the service knowledge graph, it helps to improve the accuracy of the service problem attribution model in determining the language expression logic of the text data to be analyzed. The accuracy of the service problem attribution model in matching the service problem classification results of the text data to be analyzed is also improved.
  • the first embodiment of the present application provides a service problem attribution method.
  • the specific process is shown in FIG. 8 , which is a flow chart of the service problem attribution method provided in the first embodiment of the present application.
  • the service problem attribution method shown in Figure 8 includes: step S801 to step S802.
  • step S801 text data to be analyzed for service information is obtained.
  • This step is used to obtain text data to be analyzed for service information.
  • the server obtains the text data to be analyzed, it is the data basis for the service problem attribution model to determine the service problem classification results corresponding to the text data to be analyzed in subsequent steps.
  • the service information may be food service information for food services provided by merchants to users;
  • the analyzed text data for service information may be user evaluation information for the food service information; and the obtained information for the service information
  • the text data to be analyzed includes: obtaining the text data to be analyzed for food service information sent by the client.
  • the service information takes the meals provided by the merchant for the user as an example, and the text data to be analyzed for the service information is the user's evaluation information of the meals he or she has eaten.
  • step S802 the text data to be analyzed is input into the service problem attribution model, and a service problem classification result for the text data to be analyzed is obtained.
  • This step is used to obtain the service problem classification results corresponding to the text data to be analyzed based on the service problem attribution model.
  • the following describes a first method in which the service problem attribution model obtains service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed.
  • the service problem attribution model is used to obtain a service problem classification result for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed.
  • the service knowledge graph helps to understand the language expression logic of the text data to be analyzed and improves the accuracy of analyzing the language expression logic of the text data to be analyzed. Therefore, the text data to be analyzed is input into the service problem attribution model.
  • the service problem attribution model determines the service problem classification results corresponding to the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model to obtain service problem classification results for the text data to be analyzed includes:
  • Step 1-1 Input the text data to be analyzed into the service problem attribution model, and obtain the first embedding vector for the text data to be analyzed.
  • the first embedding vector includes the text data embedding vector to be analyzed and The service knowledge graph embedding vector related to the text data to be analyzed;
  • Step 1-2 According to the first embedding vector, obtain the service problem classification result corresponding to the text data to be analyzed.
  • step 1-1 the text data to be analyzed is input into the service problem attribution model to obtain Obtaining the first embedding vector for the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model to obtain the text data embedding vector to be analyzed; embedding the text data to be analyzed according to the Vector, query the service knowledge graph embedding vector associated with the to-be-analyzed text data embedding vector; according to the to-be-analyzed text data embedding vector and the service knowledge graph embedding vector, obtain the first for the to-be-analyzed text data embedding vector.
  • the service problem attribution model obtains the text data to be analyzed, obtains the service knowledge graph related to the text data to be analyzed, uses the text data to be analyzed and the service knowledge graph together as input information, and analyzes and determines the service of the text data to be analyzed. Problem classification results.
  • the service problem attribution model converts the data form of the text data to be analyzed into the text data embedding vector to be analyzed, and converts the data form of the service knowledge graph into the service knowledge graph embedding vector.
  • a first embedding vector for the text data to be analyzed is generated according to the embedding vector of the text data to be analyzed and the service knowledge graph embedding vector related to the text data to be analyzed.
  • obtaining the service knowledge graph related to the text to be analyzed may be to query the service knowledge graph related to the text data to be analyzed from the service knowledge graph library pre-stored by the service problem attribution model, or to obtain the service knowledge graph related to the text data to be analyzed from other sharing platforms.
  • the service knowledge graph related to the text data to be analyzed may be to query the service knowledge graph related to the text data to be analyzed from the service knowledge graph library pre-stored by the service problem attribution model, or to obtain the service knowledge graph related to the text data to be analyzed from other sharing platforms.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining an embedding vector of the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model, Obtain the text embedding vector corresponding to each text unit in the text data to be analyzed, the paragraph embedding vector corresponding to each text unit in the text data to be analyzed, and the corresponding paragraph embedding vector of each text unit in the text data to be analyzed.
  • Position embedding vector embedding the text vector corresponding to each text unit in the text data to be analyzed, the paragraph embedding vector corresponding to each text unit in the text data to be analyzed, and the embedding vector of each text unit in the text data to be analyzed.
  • the corresponding position embedding vector in the text data is processed to obtain the embedding vector of the text data to be analyzed.
  • querying the service knowledge graph embedding vector associated with the text data embedding vector to be analyzed according to the embedding vector of the text data to be analyzed includes: obtaining the embedding vector of the text data to be analyzed according to the embedding vector of the text data to be analyzed.
  • the service information embedding vector in the text data query the service knowledge graph for the service information according to the service information embedding vector in the text data to be analyzed; obtain the correspondence of each text unit in the service knowledge graph for the service information
  • the text embedding vector of The text embedding vector corresponding to each text unit, the paragraph embedding vector corresponding to each text unit in the service knowledge graph, and the corresponding position embedding vector of each text unit in the service knowledge graph are processed to obtain the Service knowledge graph embedding vector of service information.
  • step 1-2 according to the first embedding vector, the service problem classification result corresponding to the text data to be analyzed is obtained, which can be obtained in at least the following three ways, which are discussed separately below.
  • the characteristic information of the text data to be analyzed is obtained; according to the characteristic information of the text data to be analyzed, the service problem classification list of the service problem attribution model is queried with the characteristics of the text data to be analyzed.
  • the target service problem classification result matched by the feature information of the text data is used as the service problem classification result corresponding to the text data to be analyzed.
  • the target service problem classification result matching the characteristic information of the text data to be analyzed is queried in the service problem classification list of the service problem attribution model, as the result of the classification of the target service problem.
  • the service problem classification results corresponding to the text data to be analyzed include: obtaining the characteristic information corresponding to multiple candidate service problem classification results in the service problem classification list of the service problem attribution model; and converting the characteristics of the text data to be analyzed.
  • the information is compared with the feature information respectively corresponding to the plurality of candidate service question classification results, and the candidate service question classification result containing the feature information of the text data to be analyzed is determined to match the feature information of the text data to be analyzed.
  • the target service problem classification result is used as the service problem classification result corresponding to the text data to be analyzed.
  • the text data to be analyzed is "There are mold spots on the fruits in this fruit drink", and the characteristic information of the text data to be analyzed is determined to be "fruit and mold spots" according to the first embedding vector.
  • the first embedding vector includes the text data embedding vector to be analyzed and the service knowledge graph embedding vector.
  • the text data embedding vector to be analyzed includes a text embedding vector, a position embedding vector and a paragraph embedding vector.
  • the first embedding vector is an embedding vector obtained by combining the embedding vector of the text to be analyzed and the embedding vector of the service knowledge graph. Therefore, to analyze the characteristic information of the text data to be analyzed based on the first embedding vector, it is necessary to determine the comprehensive semantics of the text data to be analyzed by combining the semantics of the text data to be analyzed and the service knowledge graph corresponding to the text data to be analyzed.
  • the service knowledge map corresponding to the text data to be analyzed "There are mold spots on the fruits in this fruit drink” is "Fruit belongs to fresh fruits and vegetables, and mold spots on the surface of fresh fruits and vegetables is a phenomenon of mildew.” Therefore, according to the first It can be seen from the embedding vector that the characteristic information of the text data to be analyzed is "fruit and mold spots”.
  • the classification results of multiple candidate service problems in the service problem classification list are the two candidate service problem classification results of "visible moldy deterioration” and "rot” as shown in Figure 2.
  • the characteristic information of "visible moldy deterioration” includes “mildew characteristics appear on the surface of the object”
  • the characteristic information of "rot” includes "a large number of lesions appear inside the object.” Comparing the characteristic information of the text data to be analyzed with the characteristic information of the two candidate service problem classification results mentioned above, it is obtained that the service problem classification result corresponding to the text data to be analyzed "There are mold spots on the fruits in this fruit drink" is visible mold. Go bad.
  • the characteristic information of the text data to be analyzed is obtained; according to the characteristic information of the text data to be analyzed, the service problem classification list of the service problem attribution model is queried with the characteristics of the text data to be analyzed. At least one service problem classification result matching the feature information of the text data; the at least one service problem classification result is used as the service problem classification result corresponding to the text data to be analyzed.
  • the characteristic information of the text data to be analyzed is obtained; according to the characteristic information of the text data to be analyzed, the service problem classification list in the service problem attribution model is queried with the The service problem classification result associated with the characteristic information of the text data to be analyzed is used as the target service problem classification result of the text data to be analyzed; on the basis of determining the target service problem classification result of the text data to be analyzed, Query the small classification results of target service problems associated with the characteristic information of the text data to be analyzed in the small classification list of service problems of the large classification result of the target service problem; according to the large classification result of the target service problem and the target Service problem classification results are generated to generate service problem classification results corresponding to the text data to be analyzed.
  • the service problem attribution model includes a large classification text data feature analysis model, and the large classification text data feature analysis model is used to analyze the feature information of text data corresponding to the service problem large classification results; Analyze the characteristic information of the text data, and query the service problem classification results associated with the characteristic information of the text data to be analyzed in the service problem classification list in the service problem attribution model as the text data to be analyzed.
  • the target service problem large classification result includes: according to the large classification text data feature analysis model, obtaining the first feature information of the text data in the text data to be analyzed for analyzing the service problem large classification result; obtaining the service The first candidate feature information of the text data corresponding to the large classification result of each candidate service question in the question classification list; the first feature information in the text data to be analyzed and the first candidate feature information corresponding to the large classification result of each candidate service question The first candidate feature information is compared to obtain a large classification result of the target service problem corresponding to the text data to be analyzed.
  • the associated target service problem sub-classification results include: obtaining the second feature information of the text data used to analyze the service problem sub-classification results in the text data to be analyzed; obtaining each candidate service in the service problem sub-classification list The second candidate feature information of the text data corresponding to the problem classification result; compare the second feature information in the text data to be analyzed with the second candidate feature information corresponding to the small classification result of each candidate service question to obtain The sub-classification results of the target service issues corresponding to the text data to be analyzed.
  • the above is the first method for the service problem attribution model to obtain service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed. Based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed, the accuracy of the language expression logic of the text data to be analyzed is improved. On this basis, the accuracy of the service problem classification results corresponding to the text data to be analyzed is improved. Accuracy.
  • the first method mentioned above is based on the service knowledge graph to improve the accuracy of the language expression logic of the text data to be analyzed.
  • the service problem attribution model also includes a second way to obtain the service problem classification results corresponding to the text data to be analyzed:
  • the service problem attribution model is specifically used to obtain information about the text to be analyzed based on the text data to be analyzed, the pinyin information corresponding to the text data to be analyzed, and the service knowledge graph related to the text data to be analyzed. Service problem classification results of the data.
  • the classification results of service issues corresponding to the text data to be analyzed are analyzed, not only combined with the service knowledge graph of the text data to be analyzed, but also combined with the pinyin information of the text data to be analyzed, to improve problems such as typos in the text data to be analyzed, and also Improved the accuracy of language expression logic for analyzing text data to be analyzed.
  • the service problem classification results of analyzing text data include:
  • Step 2-1 Input the text data to be analyzed into the service problem attribution model, and obtain a second embedding vector for the text data to be analyzed, where the second embedding vector includes the pinyin of the text data to be analyzed. Embedding vectors and pinyin embedding vectors of the service knowledge graph related to the text data to be analyzed;
  • Step 2-2 According to the second embedding vector, obtain the service problem classification result corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining a second embedding vector for the text data to be analyzed includes: converting the text data to be analyzed Input it into the service problem attribution model to obtain the text data embedding vector to be analyzed, which includes the pinyin embedding vector corresponding to each text unit in the text data to be analyzed; according to the text data to be analyzed Text data embedding vector, query the service knowledge graph embedding vector associated with the text data embedding vector to be analyzed, the service knowledge graph embedding vector includes the pinyin embedding vector corresponding to each text unit in the service knowledge graph; according to the The text data to be analyzed embedding vector and the service knowledge graph embedding vector are described to obtain a second embedding vector for the text data to be analyzed.
  • the service problem attribution model converts the data form of the text data to be analyzed into the form of embedding vectors.
  • it also obtains the text unit Pinyin embedding vector.
  • the text data to be analyzed "There are cockroaches in this meal” contains the text embedding vector, paragraph embedding vector, position embedding vector and pinyin embedding vector of each word.
  • the addition of the pinyin embedding vector of the text unit can improve the probability of typos when identifying the text of the text data to be analyzed.
  • the embedding vector of the text data to be analyzed is combined with the embedding vector of the service knowledge graph related to the text data to be analyzed to generate a second embedding vector for the text data to be analyzed.
  • the pinyin embedding vector and the service knowledge graph are combined to not only improve the accuracy of character recognition in the text data to be analyzed, but also improve the accuracy of the linguistic logical expression of the text data to be analyzed.
  • the second embedding vector obtained in step 201 is obtained after vector processing by the text data embedding vector to be analyzed and the service knowledge graph embedding vector.
  • the text data embedding vector to be analyzed in this step includes the pinyin embedding vector of the text data to be analyzed. .
  • the step of inputting the text data to be analyzed into the service problem attribution model and obtaining the embedding vector of the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model and obtaining the embedding vector of the text data to be analyzed.
  • the corresponding paragraph embedding vector in the text data and the corresponding position embedding vector of each text unit in the text data to be analyzed are processed to obtain the embedding vector of the text data to be analyzed.
  • Querying the service knowledge graph embedding vector associated with the text data embedding vector to be analyzed according to the embedding vector of the text data to be analyzed includes: obtaining the text data to be analyzed according to the embedding vector of the text data to be analyzed.
  • the service information embedding vector in the text data to be analyzed query the service knowledge graph for the service information; obtain the text corresponding to each text unit in the service knowledge graph for the service information Embedding vector, the pinyin embedding vector corresponding to each text unit, the paragraph embedding vector corresponding to each text unit in the service knowledge map, the corresponding position embedding vector of each text unit in the service knowledge map; for all The text embedding vector corresponding to each text unit in the service knowledge graph of the service information, the pinyin embedding vector corresponding to each text unit, the paragraph embedding vector corresponding to each text unit in the service knowledge graph, each text unit in The corresponding position embedding vector in the service knowledge graph is processed to obtain a service knowledge graph embedding vector for the service information.
  • the service problem attribution model obtains the service problem classification results of the text data to be analyzed.
  • the service problem attribution model converts the text data to be analyzed into embedding vectors, based on the originally obtained text embedding vector, paragraph embedding vector, and position embedding vector, it combines the pinyin embedding vector corresponding to each text unit to reduce the need to Analyze typos in text data.
  • the accuracy of the language expression logic of the text data to be analyzed is improved.
  • the service problem classification results corresponding to the text data to be analyzed are obtained.
  • Embodiments of the present application provide a service problem attribution method, which includes: obtaining text data to be analyzed for service information; inputting the text data to be analyzed into a service problem attribution model, and obtaining text data to be analyzed for the text data to be analyzed. Service problem classification results; wherein the service problem attribution model is used to analyze the text data according to the text data to be analyzed and The service knowledge graph related to the text data to be analyzed is used to obtain service problem classification results for the text data to be analyzed.
  • the service problem attribution model analyzes and obtains the service problem classification results for the text data to be analyzed based on the text data to be analyzed and the service knowledge graph related to the text data to be analyzed.
  • the service problem attribution model provided in the above method uses the text data to be analyzed and the service knowledge graph related to the text data to be analyzed as input information.
  • the service knowledge graph can obtain the attribute information of the service information in the text data to be analyzed, thereby Improve the understanding of language expression logic for analyzing text data. Based on the service knowledge graph, it helps to improve the accuracy of the service problem attribution model in determining the language expression logic of the text data to be analyzed. The accuracy of the service problem attribution model in matching the service problem classification results of the text data to be analyzed is also improved.
  • the second embodiment of this application also provides a service problem attribution device. Since the device embodiments are basically similar to the embodiments corresponding to the application scenario and the first embodiment, the description is relatively simple. For relevant details, please refer to the embodiments corresponding to the application scenario and part of the description of the first embodiment.
  • the device embodiments described below are merely illustrative.
  • FIG. 9 is a schematic diagram of a service problem attribution device provided in the second embodiment of the present application.
  • a service problem attribution device provided in the second embodiment of the present application includes:
  • the first obtaining unit 901 is used to obtain text data to be analyzed for service information.
  • the second obtaining unit 902 is used to input the text data to be analyzed into a service problem attribution model and obtain a service problem classification result for the text data to be analyzed; wherein the service problem attribution model is used to calculate the service problem attribution model based on the text data to be analyzed.
  • the text data to be analyzed and the service knowledge graph related to the text data to be analyzed are used to obtain service problem classification results for the text data to be analyzed.
  • the third embodiment of this application also provides another service problem attribution method. .
  • FIG. 10 is a schematic diagram of another service problem attribution method provided in the third embodiment of the present application.
  • the service problem attribution method shown in Figure 10 includes: step S1001 to step S1002.
  • step S1001 text data to be analyzed for service information is obtained.
  • step S1002 the text data to be analyzed is input into the service problem attribution model, and a service problem classification result for the text data to be analyzed is obtained.
  • the service problem attribution model is used to obtain a service problem classification result for the text data to be analyzed based on the text data to be analyzed and the pinyin information corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into a service problem attribution model to obtain a service problem classification result for the text data to be analyzed includes: inputting the text data to be analyzed into a service problem attribution model.
  • an embedding vector of the text data to be analyzed is obtained.
  • the embedding vector of the text data to be analyzed includes a text embedding vector corresponding to each text unit in the text data to be analyzed, a pinyin embedding vector corresponding to each text unit, and each The paragraph embedding vector corresponding to the text unit in the text data to be analyzed, and the corresponding position of each text unit in the text data to be analyzed is embedded with a vector; according to the embedding vector of the text data to be analyzed, the text to be analyzed is obtained Service problem classification results corresponding to the data.
  • the input of the text data to be analyzed into the service problem attribution model to obtain the embedding vector of the text data to be analyzed includes: inputting the text data to be analyzed into the service problem attribution model, and The text data to be analyzed is subjected to vectorization processing to obtain the text embedding vector corresponding to each text unit in the text data to be analyzed, and the pinyin embedding vector corresponding to each text unit. Each text unit is in the text data to be analyzed.
  • the corresponding paragraph embedding vector, the corresponding position embedding vector of each text unit in the text data to be analyzed; the text embedding vector corresponding to each text unit in the text data to be analyzed, the pinyin embedding corresponding to each text unit Vector, the corresponding paragraph embedding vector of each text unit in the text data to be analyzed, and the corresponding position embedding vector of each text unit in the text data to be analyzed are processed to obtain the embedding vector of the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the embedding vector of the text data to be analyzed includes: obtaining the service problem classification result of the text data to be analyzed according to the embedding vector of the text data to be analyzed.
  • Feature information according to the feature information of the text data to be analyzed, query the target service problem classification result that matches the feature information of the text data to be analyzed in the service problem classification list of the service problem attribution model, as the The service problem classification results corresponding to the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the embedding vector of the text data to be analyzed includes: obtaining the service problem classification result of the text data to be analyzed according to the embedding vector of the text data to be analyzed.
  • Characteristic information according to the characteristic information of the text data to be analyzed, query the service problem classification results associated with the characteristic information of the text data to be analyzed in the service problem classification list in the service problem attribution model, As the large classification result of the target service issue of the text data to be analyzed; on the basis of determining the large classification result of the target service issue of the text data to be analyzed, in the small classification list of service issues of the large classification result of the target service issue Query the small classification results of target service questions associated with the characteristic information of the text data to be analyzed; generate services corresponding to the text data to be analyzed based on the large classification results of the target service questions and the small classification results of the target service questions Problem classification results.
  • the service problem attribution model is specifically used to obtain the service knowledge map for the text data to be analyzed based on the text data to be analyzed, the pinyin information corresponding to the text data to be analyzed, and the service knowledge graph related to the text data to be analyzed. Describe the service problem classification results of the text data to be analyzed.
  • inputting the text data to be analyzed into a service problem attribution model to obtain a service problem classification result for the text data to be analyzed includes: inputting the text data to be analyzed into a service problem attribution model.
  • a second embedding vector for the text data to be analyzed is obtained, the second embedding vector includes an embedding vector of the text data to be analyzed and a service knowledge graph embedding vector related to the text data to be analyzed; according to the The second embedding vector obtains the service problem classification result corresponding to the text data to be analyzed.
  • inputting the text data to be analyzed into the service problem attribution model and obtaining a second embedding vector for the text data to be analyzed includes: inputting the text data to be analyzed into the service In the problem attribution model, an embedding vector of the text data to be analyzed is obtained.
  • the embedding vector of the text data to be analyzed includes the pinyin embedding vector corresponding to each text unit in the text data to be analyzed.
  • the service knowledge graph embedding vector includes the pinyin embedding vector corresponding to each text unit in the service knowledge graph; according to the text data to be analyzed The embedding vector and the service knowledge graph embedding vector are used to obtain a second embedding vector for the text data to be analyzed.
  • obtaining the service problem classification result corresponding to the text data to be analyzed according to the second embedding vector includes: obtaining the characteristic information of the text data to be analyzed according to the second embedding vector; For the characteristic information of the text data to be analyzed, query the service problem classification results associated with the characteristic information of the text data to be analyzed in the service problem classification list in the service problem attribution model.
  • the fourth embodiment of the present application also provides another service problem attribution device.
  • the device embodiment is basically similar to the embodiment corresponding to the application scenario and the third embodiment, the description is relatively simple. For relevant details, please refer to the embodiment corresponding to the application scenario and part of the description of the third embodiment.
  • the device embodiments described below are merely illustrative.
  • FIG. 11 is a schematic diagram of another service problem attribution device provided in the fourth embodiment of the present application.
  • Another service problem attribution device provided in the fourth embodiment of the present application includes:
  • the third obtaining unit 1101 is used to obtain text data to be analyzed for service information.
  • the fourth obtaining unit 1102 is configured to input the text data to be analyzed into the service problem attribution model and obtain the service problem classification results for the text data to be analyzed.
  • the service problem attribution model is used to obtain a service problem classification result for the text data to be analyzed based on the text data to be analyzed and the pinyin information corresponding to the text data to be analyzed.
  • the fifth embodiment of this application also provides a service problem attribution model. Training methods.
  • FIG. 12 is a schematic diagram of a training method for a service problem attribution model provided in the fifth embodiment of the present application.
  • the service problem attribution method shown in Figure 12 includes: step S1201 to step S1203.
  • step S1201 a pre-trained model for analyzing feature information of text data is obtained.
  • step S1202 the pre-training model is adjusted according to the text data sample for service information and the service problem large classification result sample for the text data sample, and a large classification text data feature analysis is obtained A model, the large classification text data feature analysis model is used to analyze the feature information of the text data corresponding to the large classification result of the service problem.
  • step S1203 the large classification text data feature analysis model is adjusted based on the text data sample for service information and the service problem small classification result sample for the text data sample to obtain A service problem attribution model that analyzes the service problem classification results for the text data to be analyzed.
  • the pre-trained model analyzes the feature information of the text data in the following manner: obtains a text data embedding vector according to the text data, and the text data embedding vector includes the text embedding corresponding to each text unit in the text data. vector, the pinyin embedding vector corresponding to each text unit, the paragraph embedding vector of each text unit in the text data, and the position embedding vector of each text unit in the text data; embedding vectors according to the text data , analyze the characteristic information of the text data.
  • the method further includes: obtaining a service knowledge graph embedding vector associated with the text data according to the text data embedding vector, where the service knowledge graph embedding vector includes each of the service knowledge graph embedding vectors associated with the text data.
  • adjusting the pre-training model based on text data samples for service information and large classification result samples of service problems for the text data samples to obtain a large classification text data feature analysis model includes: The text data sample for the service information is input into the pre-training model, and the first service question large classification result of the text data sample for the service information output by the pre-training model is obtained; according to the first service question The degree of similarity between the large classification result and the service problem large classification result sample for the text data sample is adjusted to the service problem large classification parameters of the pre-training model to obtain the large classification text data feature analysis model .
  • the text data sample for the service information is input into the pre-training model, and the first service problem classification result of the text data sample for the service information output by the pre-training model is obtained
  • the method includes: inputting the text data sample for service information into the pre-training model to obtain characteristic information of the text data sample; and obtaining, according to the characteristic information of the text data sample, the pre-training model output for the pre-training model.
  • the large classification text data feature analysis model is adjusted based on the text data samples for service information and the small classification result samples of service problems for the text data samples to obtain the analysis model for the text to be analyzed.
  • the service problem attribution model of the service problem classification result of the data includes: inputting the text data for the service information into the large classification text data feature analysis model, and obtaining the large classification text data feature analysis model output for all the service problem attribution models.
  • the second service problem small classification result of the text data sample according to the similarity between the second service problem small classification result and the service problem small classification result sample, adjust the large classification text data feature analysis model , obtain the service problem attribution model used to analyze the service problem classification results for the text data to be analyzed.
  • it also includes: using text data samples for service information and service problem classification result samples that do not belong to the text data as the first negative sample pair, adjusting the pre-training model to obtain corresponding data for analysis A large classification text data feature analysis model based on the characteristic information of the text data of the service problem large classification result; using the text data sample for the service information and the service problem small classification result sample that does not belong to the text data as the second negative sample pair, The large classification text data feature analysis model is adjusted to obtain a service problem attribution model used to analyze the service problem classification results of the text data to be analyzed.
  • the sixth embodiment of the present application also provides a service problem attribution method. Because of the model training device. Since the device embodiment is basically similar to the embodiment corresponding to the application scenario and the fifth embodiment, the description is relatively simple. For relevant details, please refer to the embodiment corresponding to the application scenario and part of the description of the fifth embodiment.
  • the device embodiments described below are merely illustrative.
  • FIG. 13 is a schematic diagram of a training device for a service problem attribution model provided in the sixth embodiment of the present application.
  • a training device for a service problem attribution model provided in the sixth embodiment of the present application includes:
  • the pre-trained model obtaining unit 1301 is used to obtain a pre-trained model for analyzing feature information of text data.
  • the large classification text data feature analysis model acquisition unit 1302 is used to adjust the pre-training model according to the text data samples for service information and the service problem large classification result samples for the text data samples to obtain large classification text data.
  • Feature analysis model the large classification text data feature analysis model is used to analyze the feature information of the text data corresponding to the large classification result of the service problem.
  • the service problem attribution model obtaining unit 1303 is used to adjust the large classification text data feature analysis model according to the text data sample for the service information and the service problem small classification result sample for the text data sample, and obtain the service problem attribution model for A service problem attribution model that analyzes the service problem classification results for the text data to be analyzed.
  • a seventh embodiment of this application also provides an electronic device. Since the seventh embodiment is basically similar to the above-mentioned method embodiment provided in this application, the description is relatively simple. For relevant details, please refer to the description of the above-mentioned method embodiment provided in this application.
  • the seventh embodiment described below is merely illustrative.
  • FIG. 14 is a schematic diagram of an electronic device provided in the seventh embodiment of the present application.
  • the electronic device includes: at least one processor 1401, at least one communication interface 1402, at least one memory 1403 and at least one communication bus 1404; optionally, the communication interface 1402 can be an interface of a communication module, such as WLAN (Wireless Local Area Network) , wireless LAN) module interface; the processor 140l may be a processor CPU (Central Processing Unit, central processing unit), or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or be configured to implement the embodiment of the present application One or more integrated circuits.
  • a communication module such as WLAN (Wireless Local Area Network) , wireless LAN) module interface
  • the processor 140l may be a processor CPU (Central Processing Unit, central processing unit), or a specific integrated circuit ASIC (Application Specific Integrated Circuit), or be configured to implement the embodiment of the present application One or more integrated circuits.
  • CPU Central Processing Unit, central processing unit
  • ASIC Application Specific Integrated Circuit
  • the memory 1403 may include high-speed RAM (Random Access Memory), and may also include non-volatile memory (non-volatile memory), such as at least one disk memory.
  • the memory 1403 stores a program, and the processor 1401 calls the program stored in the memory 1403 to execute the above method provided by the embodiment of the present application.
  • the eighth embodiment of this application also provides a computer storage medium. Since the eighth embodiment is basically similar to the above method embodiments provided by this application, the description is relatively simple. , for relevant information, please refer to the description of the above method embodiment provided in this application.
  • the eighth embodiment described below is merely illustrative.
  • the computer storage medium stores a computer program, and when the program is executed, the method provided in the above method embodiment is implemented. It should be noted that for a detailed description of the storage medium provided by the eighth embodiment of the present application, reference may be made to the relevant description of the above method embodiment provided by the present application, and will not be described again here.
  • a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • Memory may include non-permanent storage in computer-readable media, random access memory (RAM) and/or non-volatile memory in the form of read-only memory (Read-Only Memory, ROM) or flash memory (f;ash). RAM).
  • RAM random access memory
  • ROM read-only memory
  • f;ash flash memory
  • Memory is an example of computer-readable media.
  • Computer-readable media includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology. Information may be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change RAM (PRAM), static random access memory (Static RAM, SRAM), dynamic random access memory (Dynamic RAM, DRAM), other types of Random access memory (RAM), read-only memory (ROM), electrically erasable programmable ROM (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD) -ROM), Digital Versatile Disc (DVD) or other optical storage, magnetic tape cassettes, tape magnetic disk storage or other magnetic storage devices or any other non-transmission medium that can be used to store information that can be accessed by a computing device. As defined in this article, computer-readable media does not include non-transitory computer-readable media (Transitory Media), such as modulated data signals and carrier waves.
  • Transitory Media such as

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Health & Medical Sciences (AREA)
  • Animal Behavior & Ethology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本申请实施例提供一种服务问题归因方法及装置,该方法采用服务问题归因模型分析待分析文本数据对应的服务问题分类结果。服务问题归因模型根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,分析获得针对待分析文本数据的服务问题分类结果。上述方法中提供的服务问题归因模型,将待分析文本数据和与待分析文本数据相关的服务知识图谱一起作为输入信息,服务知识图谱能获取待分析文本数据中的服务信息的属性信息,从而提升对待分析文本数据的语言表达逻辑的理解度。基于服务知识图谱有助于提升服务问题归因模型确定待分析文本数据的语言表达逻辑的准确度,服务问题归因模型对待分析文本数据匹配服务问题分类结果的准确度也得到提升。

Description

一种服务问题归因方法及装置 技术领域
本申请涉及计算机技术领域,具体涉及服务问题归因方法和装置、电子设备及计算机存储介质,服务问题归因模型的训练方法和装置、电子设备及计算机存储介质。
背景技术
目前,商家在线上平台为用户提供各种服务,用户使用该服务后,在线上平台针对该服务进行评价。线上平台为了快速分析用户提供的针对服务信息的评价信息中存在的问题,使用分析模型分析用户提供的评价信息,识别该评价信息对应的问题类型。
现有技术中,上述分析商户服务信息的问题类型的方法通常存在识别结果错误,识别准确度偏低的问题。因此,如何提升分析评价信息中的问题类型的准确率是需要解决的问题。
发明内容
本申请实施例提供一种服务问题归因方法,包括:获得针对服务信息的待分析文本数据;将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第一嵌入向量,所述第一嵌入向量包括待分析文本数据嵌入向量和与所述待分析文本数据相关的服务知识图谱嵌入向量;根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第一嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量;根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量;根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第一嵌入向量。
可选的,所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
可选的,所述根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果,包括:获取所述服务问题归因模型的服务问题分类列表中多个候选服务问题分类结果分别对应的特征信息;将所述待分析文本数据的特征信息与所述多个候选服务问题分类结果分别对应的特征信息进行比较,将包含所述待分析文本数据的特征信息的候选服务问题分类结果确定为与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
可选的,所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的至少一种服务问题分类结果;将所述至少一种服务问题分类结果作为所述待分析文本数据对应的服务问题分类结果。
可选的,所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分 类结果,包括:根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题第一分类结果,作为所述待分析文本数据的目标服务问题第一分类结果;在确定所述待分析文本数据的目标服务问题第一分类结果的基础上,在所述目标服务问题第一分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题第二分类结果;根据所述目标服务问题第一分类结果和所述目标服务问题第二分类结果,生成所述待分析文本数据对应的服务问题分类结果。
可选的,所述服务问题归因模型包括大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题第一分类结果的文本数据的特征信息;所述根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题第一分类结果,作为所述待分析文本数据的目标服务问题第一分类结果,包括:根据所述大分类文本数据特征分析模型,获取所述待分析文本数据中用于分析服务问题第一分类结果的文本数据的第一特征信息;获取所述服务问题大分类列表中每个候选服务问题第一分类结果对应的文本数据的第一候选特征信息;将所述待分析文本数据中的第一特征信息与所述每个候选服务问题第一分类结果对应的第一候选特征信息进行比较,获得所述待分析文本数据对应的目标服务问题第一分类结果。
可选的,所述在确定所述待分析文本数据的目标服务问题第一分类结果的基础上,在所述目标服务问题第一分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题第二分类结果,包括:获取所述待分析文本数据中用于分析服务问题第二分类结果的文本数据的第二特征信息;获取所述服务问题小分类列表中每个候选服务问题第二分类结果对应的文本数据的第二候选特征信息;将所述待分析文本数据中的第二特征信息与所述每个候选服务问题第二分类结果对应的第二候选特征信息进行比较,获得所述待分析文本数据对应的目标服务问题第二分类结果。
可选的,所述将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;将所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,以及每个文本单元在所述待分析文本数据中对应的位置嵌入向量进行处理,获得所述待分析文本数据嵌入向量。
可选的,所述根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据中的服务信息嵌入向量;根据所述待分析文本数据中的服务信息嵌入向量,查询针对所述服务信息的服务知识图谱;获取针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量;将针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量进行处理,获得针对所述服务信息的服务知识图谱嵌入向量。
可选的,所述服务信息为商户向用户提供的针对食品服务的食品服务信息;所述针对服务信息的分析文本数据为用户针对所述食品服务信息的评价信息;所述获得针对服务信息的待分析文本数据,包括:获得用户端发送的针对食品服务信息的待分析文本数据。
可选的,所述服务问题归因模型具体用于根据所述待分析文本数据,所述待分析文本数据对应的拼音信息,以及与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,所述第二嵌入向量包括所述待分析文本数据的拼音嵌入向量和与所述待分析文本数据相关的服务知识图谱的拼音嵌入向量;根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析 文本数据中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括所述服务知识图谱中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第二嵌入向量。
可选的,所述将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,包括:将所述待分析文本数据输入所述服务问题归因模型中,获得所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;将所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,以及每个文本单元在所述待分析文本数据中对应的位置嵌入向量进行处理,获得所述待分析文本数据嵌入向量。
可选的,所述根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据中的服务信息嵌入向量;根据所述待分析文本数据中的服务信息嵌入向量,查询针对所述服务信息的服务知识图谱;获取针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量;将针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量进行处理,获得针对所述服务信息的服务知识图谱嵌入向量。
本申请实施例还提供一种服务问题归因方法,包括:获得针对服务信息的待分析文本数据;将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据以及所述待分析文本数据对应的拼音信息,获得针对所述待分析文本数据的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:将所述待分析文本数据输入到服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得待分析文本数据嵌入向量,包括:将所述待分析文本数据输入到服务问题归因模型中,对所述待分析文本数据进行向量化处理,获得所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;对所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量进行处理,获得所述待分析文本数据嵌入向量。
可选的,所述根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
可选的,所述根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题第一分类结果,作为所述待分析文本数据的目标服务问题第一分类结果;在确定所述待分析文本数据的目标服务问题第一分类结果的基础上,在所述目标服务问题第一分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题第二分类结果;根据所述目标服务问题第一分类结果和所述目标服务问题第二分类结果,生成所述待分析文本数据对应的服务问题分类结果。
可选的,所述服务问题归因模型具体用于根据所述待分析文本数据,所述待分析文本数据对应的拼音信息,以及与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,所述第二嵌入向量包括待分析文本数据嵌入向量和与所述待分析文本数据相关的服务知识图谱嵌入向量;根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析文本数据中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括所述服务知识图谱中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第二嵌入向量。
可选的,所述根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述第二嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题第一分类结果,作为所述待分析文本数据的目标服务问题第一分类结果;在确定所述待分析文本数据的目标服务问题第一分类结果的基础上,在所述目标服务问题第一分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题第二分类结果;根据所述目标服务问题第一分类结果和所述目标服务问题第二分类结果,生成所述待分析文本数据对应的服务问题分类结果。
本申请实施例还提供一种服务问题归因模型的训练方法,包括:获得用于分析文本数据的特征信息的预训练模型;根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第一分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题第一分类结果的文本数据的特征信息;根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第二分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
可选的,所述预训练模型通过如下方式分析文本数据的特征信息:根据所述文本数据获得文本数据嵌入向量,所述文本数据嵌入向量包括所述文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述文本数据中的段落嵌入向量,以及每个文本单元在所述文本数据中的位置嵌入向量;根据所述文本数据嵌入向量,分析所述文本数据的特征信息。
可选的,还包括:根据所述文本数据嵌入向量获取与所述文本数据相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括与所述文本数据相关联的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中的段落嵌入向量,以及每个文本单元在所述服务知识图谱中的位置嵌入向量;所述根据所述文本数据嵌入向量,分析所述文本数据的特征信息,包括:根据所述文本数据嵌入向量,以及与所述文本数据相关联的服务知识图谱嵌入向量,分析所述文本数据的特征信息。
可选的,所述根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第一分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,包括:将所述针对服务信息的文本数据样本输入所述预训练模型中,获得所述预训练模型输出的针对所述服务信息的文本数据样本的第一服务问题第一分类结果;根据所述第一服务问题第一分类结果和所述针对所述文本数据样本的服务问题第一分类结果样本之间的相似程度,对所述预训练模型进行服务问题大分类参数的调整,获得所述大分类文本数据特征分析模型。
可选的,所述将所述针对服务信息的文本数据样本输入所述预训练模型中,获得所述预训练模型输出的针对所述服务信息的文本数据样本的第一服务问题第一分类结果,包括:将所述针对服务信息的文本数据样本输入所述预训练模型中,获得所述文本数据样本的特征信息;根据所述文本数据样本的特征信息,获得所述预训练模型输出的针对所述服务信息的文本数据样本的第一服务问题第一分类结果。
可选的,所述根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第二分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型,包括:将针对所述服务信息的文本数据输入所述大分类文本数据特征分析模型中,获得所述大分类文本数据特征分析模型输出的针对所述文本数据样本的第二服务问题第二分类结果;根据所述第二服务问题第二分类结果与所述服务问题第二分类结果样本之间的相似程度,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
可选的,还包括:将针对服务信息的文本数据样本和不属于所述文本数据的服务问题第一分类结果样本作为第一负样本对,对所述预训练模型进行调整,获得用于分析对应于服务问题第一分类结果的文本数据的特征信息的大分类文本数据特征分析模型;将针对服务信息的文本数据样本和不属于所述文本数据的服务问题第二分类结果样本作为第二负样本对,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
本申请实施例还提供一种服务问题归因装置,包括:第一获得单元,用于获得针对服务信息的待分析文本数据;第二获得单元,用于将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
本申请实施例还提供一种服务问题归因装置,包括:第三获得单元,用于获得针对服务信息的待分析文本数据;第四获得单元,用于将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据以及所述待分析文本数据对应的拼音信息,获得针对所述待分析文本数据的服务问题分类结果。
本申请实施例还提供一种服务问题归因模型的训练装置,包括:预训练模型获得单元,用于获得用于分析文本数据的特征信息的预训练模型;大分类文本数据特征分析模型获得单元,用于根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第一分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题第一分类结果的文本数据的特征信息;服务问题归因模型获得单元,用于根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第二分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
本申请实施例还提供一种电子设备,所述电子设备包括处理器和存储器;所述存储器中存储有计算机程序,所述处理器运行所述计算机程序后,执行上述方法。
本申请实施例还提供一种计算机存储介质,所述计算机存储介质存储有计算机程序,所述计算机程序被执行时执行上述方法。
与现有技术相比,本申请实施例具有如下优点:
本申请实施例提供一种服务问题归因方法,包括:获得针对服务信息的待分析文本数据;将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
在本申请的一个实施例中,服务问题归因模型根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,分析获得针对待分析文本数据的服务问题分类结果。上述方法中提供的服务问题归因模型,将待分析文本数据和与待分析文本数据相关的服务知识图谱一起作为输入信息,服务知识图谱能获取待分析文本数据中的服务信息的属性信息,从而提升对待分析文本数据的语言表达逻辑的理解度。基于服务知识图谱有助于提升服务问题归因模型确定待分析文本数据的语言表达逻辑的准确度,服务问题归因模型对待分析文本数据匹配服务问题分类结果的准确度也得到提升。
在本申请的一个实施例中,服务问题归因模型根据待分析文本数据,和待分析文本数据对应的拼音信息,分析获得针对待分析文本数据的服务问题分类结果。上述方法中提供的服务问题归因模型,在对待分析文本数据进行处理时,结合待分析文本数据的拼音信息,提升待分析文本数据的语言表达逻辑的准确度。基于待分析文本数据有助于提升待分析文本数据的语言表达逻辑的准确度,服务问题归因模型对待分析文本数据匹配 服务问题分类结果的准确度也得到提升。
在本申请的一个实施例中,其提供了一种服务问题归因模型的训练方法,包括:获得用于分析文本数据的特征信息的预训练模型;根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第一分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题第一分类结果的文本数据的特征信息;根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第二分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
上述方法,获得预训练模型,首先采用文本数据样本和针对待分析文本数据样本的服务问题第一分类结果样本,对预训练模型进行调整,获得大分类文本数据特征分析模型。然后,根据文本数据样本和针对文本数据样本的服务问题第二分类结果样本,对大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。由此可知,训练好的服务问题归因模型获取待分析文本数据的服务问题分类结果时,先获取待分析文本数据的服务问题第一分类结果,在确定待分析文本数据的服务问题第一分类结果的基础上,在目标服务问题第一分类结果对应的多个服务问题第二分类结果中,确定待分析文本数据的服务问题第二分类结果。因此,最终获得的待分析文本数据的服务问题分类结果包括服务问题第一分类结果和服务问题第二分类结果。
附图说明
图1为本申请实施例提供的服务问题归因方法的应用场景图。
图2为本申请实施例提供的服务问题归因模型中针对文本信息的服务问题分类结果示意图。
图3为本申请实施例提供的商户端体检中心的展示页面。
图4为本申请实施例提供的商户店铺的食品安全事件明细信息。
图5为图4的食品安全事件明细信息的数据信息。
图6为图5中的商户发生的食品安全事件等级以及该食品安全事件对应的处理方式。
图7为本申请实施例提供的为商户制定的针对食品安全事件的学习方法。
图8为本申请第一实施例中提供的一种服务问题归因方法的流程图。
图9为本申请第二实施例中提供的一种服务问题归因装置的示意图。
图10为本申请第三实施例中提供的另一种服务问题归因方法的示意图。
图11为本申请第四实施例中提供的另一种服务问题归因装置的示意图。
图12为本申请第五实施例中提供的一种服务问题归因模型的训练方法的示意图。
图13为本申请第六实施例中提供的一种服务问题归因模型的训练装置的示意图。
图14为本申请第七实施例中提供的一种电子设备的示意图。
具体实施方式
在下面的描述中阐述了很多具体细节以便于充分理解本申请。但是本申请能够以很多示同于在此描述的其它方式来实施,本领域技术人员可以在不违背本申请内涵的情况下做类似推广,因此本申请不受下面公开的具体实施例的限制。
本申请中使用的术语是仅仅出于对特定实施例描述的目的,而非旨在限制本申请。在本申请中和所附权利要求书中所使用的描述方式例如:“一种”、“第一”、和“第二”等,并非对数量上的限定或先后顺序上的限定,而是用来将同一类型的信息彼此区分。
为了更清楚的展示本申请提供的服务问题归因方法,先介绍一下本申请提供的服务问题归因方法的应用场景。
本申请提供的服务问题归因方法,可以应用于食品安全问题分类场景。例如,线上到线下餐饮服务系统为用户提供餐品订购服务。以图1为例,其为本申请实施例提供的服务问题归因方法的应用场景图。用户通过用户端101订购餐品,退订餐品,以及用户反馈针对餐品的相关数据,例如,针对退订餐品的评价信息,食品安全险理赔评价信息,以及针对食用过的餐品的评价信息等。其中,评价信息中包括用户针对部分食品的安全问题的投诉信息。服务端102接收到用户端发送的针对该餐品的评价信息,根据该评价信息,判断该评价信息是否为针对餐品的食品安全问题进行的投诉信息,如果是,则确定该评价信息对应的食品安全问题类型,并将该评价信息以及针对该评价信息的食品安全问题类型,发送给提供该餐品的商家。
服务端102分析用户提供的评价信息对应的食品安全问题类型,是通过服务问题归因模型进行分析的。具体的,将获得的针对服务信息的待分析文本数据输入服务问题归因模型中,获得服务问题归因模型输出的针对待分析文本数据的服务问题分类结果。
例如,将用户对餐品的评价信息输入服务问题归因模型,服务问题归因模型对该评价信息进行分析,确定该评价信息对应的服务问题分类结果,也就是该餐品对应的食品安全问题类型。例如,用户对餐品的评价信息为:“这个水果饮品中的水果上面出现霉点了”,将该评价信息输入服务问题归因模型中,则服务问题归因模型输出的该餐品对应的食品安全问题类型为可见发霉变质。再例如,评价信息为:“这个饭菜中有蟑螂”,将该评价信息输入服务问题归因模型中,则服务问题归因模型输出的该餐品对应的食品安全问题类型为蟑螂。
服务问题归因模型用于根据用户提供的针对服务信息的评价信息,确定该评价信息对应的服务问题分类结果。具体是,先确定该评价信息对应的服务问题大分类结果(也被称为服务问题第一分类结果),在确定服务问题大分类结果的基础上,再分析该评价信息对应的服务问题小分类结果(也被称为服务问题第二分类结果)。根据服务问题大分类结果,以及服务问题大分类结果中的服务问题小分类结果,生成针对该评价信息的服务问题分类结果。以上述“这个饭菜中有蟑螂”的评价信息为例,该评价信息的服务问题大分类结果为异物,服务问题小分类结果为蟑螂,其中,蟑螂是异物中的虫鼠害异物中的小分类结果。
服务端获取用户的评价信息对应的食品安全问题类型后,向商户端发送该商户提供的食品中存在的食品安全事件,以及该食品安全事件对应的服务问题分类结果。商户通过商户端的体检中心查看其存在的食品安全事件。
如图3所示,其为本申请实施例提供的商户端体检中心的展示页面。在图3所示的页面中,商户端的体检中心展示了商户存在的安全事件统计情况,以及管控处理记录,商户的食品安全信用分数信息,以及针对食品安全问题的学习专区信息。图3的页面向商户展示了其当前存在的食品安全事件以及对应的学习策略,使得商户明确其当前存在的食品安全事件和进一步的改正方案。当商户在图3的页面中点击查看事件按钮后,进入图4所示的页面。
请参考图4,其为本申请实施例提供的商户店铺的食品安全事件的明细信息。在图4页面中,商户可以具体了解其当月时间内存在的具体食品安全事件的明细信息,商户可以根据此明细进一步改善问题。
图5为图4的食品安全事件的明细信息的数据信息。在图5中,商户能够了解其当月的食品安全事件发生率,以及食品安全事件发生次数。
图6为图5中的商户发生的食品安全事件等级以及该食品安全事件等级对应的处理方式。
同时,服务端102还会向商家所使用的商户端103提供处理该食品安全问题类型的学习课程,以供商家及时处理该食品安全问题,防止继续发生食品安全问题。
图7为本申请实施例提供的为商户制定的针对食品安全事件的学习方法。具体的,图7以虫鼠害异物防治方法为例,具体讲解防治虫鼠害异物的详情信息,对应的法规要求以及防范措施。
此外,本申请实施例提供的服务问题归因方法,还可以应用于向用户推荐餐品的场景。例如,用户登录线上外卖平台,在线上外卖平台的搜索框中搜索相关餐品关键词,线上外卖平台服务端根据用户的搜索关键词,采用对应的推荐算法,获取与搜索关键词对应的餐品信息,将获得的餐品信息发送给用户端,用户端展示获得的餐品信息。
其中,服务端根据用户提供的搜索关键词确定对应的推荐餐品时,其推荐算法参考的因素包括多个因素,例如,用户的用餐习惯信息,用户搜索的餐品的餐品特征信息, 用户所处的地理位置信息,以及提供该餐品的商户对应的食品安全等级。其中,商户的食品安全等级可以采用本申请提供的服务问题归因方法中的服务问题归因模型获得。
此外,本申请实施例提供的服务问题归因方法,还可以应用于用户端的首页所推荐的商户信息。例如,用户登录线上外卖平台后,用户端的首页展示的商户信息是由服务端预先通过推荐算法获得的。其中,推荐算法的参考因素包括:商户所处地理位置与用户所处地理位置之间的距离信息,用户的历史关注商户类型,餐品类型,以及商户的食品安全等级。其中,商户的食品安全等级可以采用本申请提供的服务问题归因方法中的服务问题归因模型获得。
其中,商家的食品安全等级可以通过如下方法获得:
用户针对该商家的外卖产品提供评价信息,采用服务问题归因模型对用户提供的评价信息进行分析,确定评价信息中的食品安全问题,以及该食品安全问题的类型。也就是说,如果用户针对其食用的餐品提供的评价信息中包含食品安全问题,则根据服务问题归因模型确定该食品安全问题对应的具体服务问题分类结果。根据确定的食品安全问题的类型确定该食品对应的商户的食品安全等级。
本申请实施例中,采用服务问题归因模型分析用户评价信息中包含的食品安全问题类型。将针对餐品的评价信息以及该评价信息中的食品安全问题对应的食品安全问题类型,发送给向用户提供该餐品的商户。一方面,提升了商户的食品安全意识,提示商户食品安全问题的严重性。另一方面,向商户提示了用户食用餐品过程中发送的具体食品安全问题。并且,向商户提供应对此类食品安全问题的方法,避免之后再次发生此类食品安全问题。其中,如果商户存在的食品安全问题的类型偏多时,向商户提供一定的强制性整改措施。
基于上述步骤,线上外卖系统可以确定商户的食品安全等级,向用户推荐对应的商户列表时,根据商户食品安全等级确定是否向用户推荐该商户。
上述各应用场景中提到的服务问题归因模型通过分析用户评价信息,获得用户评价信息对应的服务问题分类结果。其中,服务问题归因模型具体通过如下方法进行分析:
第一种,服务问题归因模型根据待分析文本数据以及与所述待分析文本数据相关联的服务知识图谱,确定针对待分析文本数据的服务问题分类结果。
其中,服务知识图谱是指待分析文本数据的先验知识图谱,其预先存储于服务问题归因模型中。
服务知识图谱至少包括如下几种类型:
1)服务知识图谱是归因服务知识图谱,也就是待分析文本数据中的服务信息的知识图谱。例如,待分析文本数据为:臭豆腐好臭。其对应的服务知识图谱为:臭豆腐是一种食品,该食品的气味属性是臭的。
采用传统的分析方法分析该待分析文本数据时,其分析结果是臭豆腐存在变质问题。
然而,本申请实施例中,服务问题归因模型分析该待分析文本数据“臭豆腐好臭”时,其会同时结合该待分析文本数据对应的服务知识图谱“臭豆腐是一种食品,该食品的气味属性是臭的”,分析该待分析文本数据对应的服务问题分类结果。该例子中,“臭豆腐好臭”和“臭豆腐是一种食品,该食品的气味属性是臭的”两者信息可以确定,该待分析文本数据中所反馈的问题不属于服务问题归因模型的归因标准中的任何一种归因类型,说明待分析文本数据所反馈的问题不属于服务问题。
服务问题归因模型对输入信息进行向量化处理,输出信息也是用向量进行表示。例如,向量为1表示该待分析文本数据的服务信息对应归因标准中的至少一种归因类型。向量为0表示该待分析文本数据的服务信息不对应归因标准中的任何一种归因类型。
再例如,待分析文本数据为:这个螺狮粉好臭。传统的分析方法中会确定螺狮粉存在变质问题。但是,本申请实施例中,服务问题归因模型分析待分析文本数据时,结合服务知识图谱。此处,待分析文本数据对应的服务知识图谱是:螺狮粉是一种商品,该商品的气味属性是臭的。因此,服务问题归因模型将待分析文本数据和其对应的服务知识图谱结合分析,得到螺狮粉的臭是螺狮粉本身的产品属性,说明待分析文本数据中所反馈的问题不属于服务问题归因模型的归因标准中的任何一种归因类型。
2)服务知识图谱是配料图谱,例如,待分析文本数据为“黑椒牛柳中怎么还有黑椒呢”,其对应的服务知识图谱为“黑椒牛柳的配料表中包括:黑椒和牛柳”。
采用传统的分析方法分析该待分析文本数据时,其分析结果是黑椒牛柳中存在异物。 但是,本申请实施例中,服务问题归因模型在分析该用户评价信息“黑椒牛柳中怎么还有黑椒呢”时,会结合其对应的服务知识图谱“黑椒牛柳的配料表中包括:黑椒和牛柳”。由此确定,黑椒牛柳中的黑椒是这道菜品的主要配料之一。因此,说明待分析文本数据所反馈的问题不属于服务问题归因模型中归因标准中的任何一种归因类型,其不存在服务问题。
3)服务知识图谱是菜谱,例如,待分析文本数据为“酸菜鱼是哪种鱼”,其对应的服务知识图谱为“酸菜鱼是菜谱,主要配料有酸菜和鱼”。
采用传统的分析方法分析待分析文本数据时,其分析结果是酸菜鱼问题是食物变质问题。但是,本申请实施例中,服务问题归因模型在分析该待分析文本数据“酸菜鱼是哪种鱼”时,会结合其对应的服务知识图谱“酸菜鱼是菜谱,主要配料有酸菜和鱼”。由此确定,酸菜鱼是一道菜,其是由普通的鱼和酸菜制作而成,而非鱼的种类名称。因此,说明待分析文本数据中所反馈的问题不属于服务问题归因模型中归因标准中的任何一种归因类型,其不存在服务问题。
4)服务知识图谱为常识图谱。例如,待分析文本数据为“耗子为之”,其对应的服务知识图谱为“好自为之”。
采用传统的分析方法分析待分析文本数据时,其分析结果是老鼠。但是,本申请实施例中,服务问题归因模型在分析待分析文本数据“耗子为之”时,会结合其对应的服务知识图谱“好自为之”。由此确定,该待分析文本数据是用户向商户提示,相对于商家之前对用户提供的餐品服务,此次服务相对较差,但不属于服务问题归因模型中归因标准中的任何一种归因类型,希望商户好自为之,改善现在的服务质量。
以上即为服务知识图谱的几种常见类型,服务问题归因模型分析待分析文本数据的语言表达逻辑时,结合待分析文本数据对应的服务知识图谱,可以提升待分析文本数据的语言表达逻辑的准确度。进一步地,根据待分析文本数据和待分析文本数据对应的服务知识图谱获得待分析文本数据的语言表达逻辑后,确定待分析文本数据对应的服务分类结果的准确度也得到提升。
服务问题归因模型采用待分析文本数据,和与待分析文本数据相关联的服务知识图谱,确定待分析文本数据对应的服务问题归因结果的过程具体如下:
将待分析文本数据输入服务问题归因模型中,获得待分析文本数据嵌入向量,该待分析文本数据嵌入向量包括待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元在所述待分析文本数据中的段落嵌入向量,每个文本单元在所述待分析文本数据中的位置嵌入向量。其中,每个文本单元可以为每个字或者每个词。如图1中,E这表示“这”这个字对应的文本嵌入向量,EA表示“这”这个字在所述待分析文本数据中的段落嵌入向量,E1表示“这”这个字在所述待分析文本数据中的位置嵌入向量。
然后,根据待分析文本数据嵌入向量,获取与待分析文本数据相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元在所述服务知识图谱中的段落嵌入向量,每个文本单元在所述服务知识图谱中的位置嵌入向量。
将待分析文本数据嵌入向量和与待分析文本数据相关联的服务知识图谱嵌入向量,生成针对待分析文本数据的第一嵌入向量。服务问题归因模型分析针对待分析文本数据的第一嵌入向量,确定待分析文本数据的特征信息用于分析服务问题分类结果,根据待分析文本数据的特征信息,确定待分析文本数据对应的服务问题分类结果。
其中,服务问题分析模型根据待分析文本数据的特征信息,确定待分析文本数据对应的服务问题分类结果。具体是,根据待分析文本数据的特征信息确定待分析文本数据对应的服务问题大分类结果,例如图2所示的异物,变质,食源性疾病,过期,餐品未熟等分类结果。其中,确定目标服务问题大分类结果(也被称为目标服务问题第一分类结果)的过程也称为固定共享层。然后,在固定共享层后,选择与待分析文本数据的特征信息对应的目标分类层,也就是确定待分析文本数据的目标服务问题小分类结果(也被称为目标服务问题第二分类结果)。例如,图2所示的大头针。其中,确定的目标服务问题小分类结果就是待分析文本数据的目标服务问题分类结果。
第二种,服务问题归因模型根据待分析文本数据以及所述待分析文本数据的拼音信息,确定针对待分析文本数据的服务问题分类结果。
具体的,将待分析文本数据输入服务问题归因模型中,获得待分析文本数据嵌入向量,待分析文本数据嵌入向量包括待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中的段落嵌入 向量,每个文本单元在所述待分析文本数据中的位置嵌入向量。
待分析文本数据嵌入向量包括每个文本单元对应的拼音嵌入向量,因此,服务问题归因模型将待分析文本数据转换为嵌入向量,能够改善错别字的现象,从而提升分析待分析文本数据的语言表达逻辑的准确度。
第三种,服务问题归因模型根据待分析文本数据、所述待分析文本数据的拼音信息、以及与所述待分析文本数据相关联的服务知识图谱,确定针对待分析文本数据的服务问题分类结果。
将待分析文本数据输入服务问题归因模型中,获得待分析文本数据嵌入向量,待分析文本数据嵌入向量包括待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中的段落嵌入向量,每个文本单元在所述待分析文本数据中的位置嵌入向量。例如图1中,E饭是饭这个字对应的文本嵌入向量,EA是饭这个字在“这个饭菜中有蟑螂”的待分析文本数据中的段落嵌入向量,E3是饭这个字在“这个饭菜中有蟑螂”的待分析文本数据中的位置嵌入向量,Efan是饭这个字对应的拼音嵌入向量。
根据待分析文本数据嵌入向量,确定与待分析文本数据相关联的服务知识图谱嵌入向量,服务知识图谱嵌入向量包括:服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中的段落嵌入向量,每个文本单元在所述服务知识图谱中的位置嵌入向量。
服务问题归因模型分析待分析文本数据时,在确定待分析文本数据的文本嵌入向量,段落嵌入向量以及位置嵌入向量的基础上,结合拼音嵌入向量,减少错别字的频率。同时,结合与待分析文本数据相关联的服务知识图谱,分析待分析文本数据的语言表达逻辑。增加拼音嵌入向量和服务知识图谱,可以提升分析待分析文本数据的语言表达逻辑的准确率,从而,服务问题归因模型可以根据待分析文本数据的语言表达逻辑提取待分析文本数据中用于分析服务问题分类结果的特征数据。根据特征数据,首先在服务问题归因模型的服务问题大分类结果列表中,确定与特征数据相对应的服务问题大分类结果,然后,基于确定的服务问题大分类结果,在服务问题大分类结果的服务问题小分类列表中选取与特征信息对应的服务问题小分类结果。根据确定的目标服务问题大分类结果和目标服务问题大分类结果中的目标服务问题小分类结果中确定待分析文本数据对应的服务问题分类结果。
此外,其中的服务问题归因模型是通过如下方法训练获得的:
第一步,获取用于分析文本数据的特征信息的预训练模型。
其中,预训练模型可以是对传统的Bert(Bidirectional Encoder Representations from Transformers,深度双向预训练编码器)模型经过训练后获得的。Bert模型为预训练的语言模型,可以用于问答系统、情感分析、垃圾邮件过滤、命名实体识别、文档聚类等任务中。
以下对Bert模型的基本概念进行介绍:
Bert模型的输入对象是待识别文本中各个字的原始向量,包括:字向量(Token Embedding,或文本向量),段落向量(Segment Embedding)和位置向量(Position Embedding),将字向量,段落向量和位置向量直接加和处理后,输入Bert模型。其中,段落向量是指使用嵌入的信息将待识别文本信息的上下文分开,位置向量是指该字在该待识别文本中的位置信息。
Bert模型包括至少三种类型的输入,并对应三种类型的输出。具体如下:
1)单文本分类:例如文章情绪分析。Bert模型在输入的段落向量中添加一个[CLS]符号向量,并将该符号对应的输出向量作为最终整篇文章的语义表示。与其他文字相比,这个[CLS]符号向量能更公平的融合各个字词的语言表达逻辑。
2)语句对文本分类:例如问答系统。其输入向量不仅添加了[CLS]符号向量,还添加了[SEP]符号向量用于分割语句。
3)序列加标注分类:此类任务每个字对应的输出向量是对该字的标注,可以理解为分类。
本申请实施例中,Bert模型中Transformer结构通过多头注意力机制和遮掩机制双向信息获取,以增强待识别文本数据的语言表达逻辑。注意力机制是让神经网络把注意力集中在一部分输入上。多头注意力机制是将每个字的多个增强语义向量进行线性组合,最终获得一个与原始字向量长度相等的增强语义向量。例如,“这个套餐/很/好吃吗” 和“这个套餐/很好/吃吗”,这两句话中,“吃”是否和“好”组合所表达的语义不同。
传统的语言模型是从左到右输入一个文本序列,或者将left-to-right和right-to-left的训练结合起来。Bert模型的双向信息获取对语境的理解对比单向的语言模型更具有优势。
遮掩机制是指将待识别文本输入模型时,随机遮盖或者替换一句话中的任意字或者任意词,让模型通过上下文的理解预测该句话中被遮盖或者替换部分的内容。然后,模型根据识别到的被遮盖或替换后的内容后,根据待识别文本的上下文内容体现出的情感,进而确定待识别文本对应的归因分类型号。
在本申请中,预训练模型相比于传统的Bert模型来说,其将待分析文本数据转换为嵌入向量时,在原始的文本嵌入向量,段落嵌入向量以及位置嵌入向量的基础上,增加拼音嵌入向量,以改善识别文字的错误率的问题。然后在待分析文本数据的嵌入向量中嵌入服务知识图谱,以提升分析待分析文本数据的语言表达逻辑的准确率。因此,预训练模型用于获取待分析文本数据的语言表达逻辑,分析确定待分析文本数据的特征信息,以供后续确定待分析文本数据的服务问题分类结果做基础。
第二步,根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题大分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题大分类结果的文本数据的特征信息。
上述预训练模型的训练目的是用于获取待分析文本数据的特征信息,此处为了使得预训练模型对服务问题大分类结果参数得到训练,采用针对服务信息的文本数据样本和针对所述文本数据样本的服务问题大分类结果样本,对预训练模型进行参数训练。
训练获得的大分类文本数据特征分析模型,根据待分析文本数据,获得待分析文本数据对应的服务问题大分类结果。其中,如图2所示,其为本申请实施例提供的服务问题归因模型中针对文本信息的服务问题分类结果示意图。服务问题大分类结果可以是异物,变质,食源性疾病,过期,餐品未熟等。
第三步,根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题小分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
所述大分类文本数据特征分析模型用于分析文本数据的服务问题大分类结果,在此基础上,对大分类文本数据特征分析模型进行进一步的参数调整,使其可以分析文本数据的服务问题小分类结果。
上述第二步中采用针对服务信息的文本数据样本和针对所述文本数据样本的服务问题大分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型。其中,此处提到的“针对所述文本数据样本的服务问题大分类结果样本”是指属于文本数据样本的服务问题大分类结果样本,其作为正样本。
此外,还可以将针对服务信息的文本数据样本和不属于所述文本数据样本的服务问题大分类结果样本作为第一负样本对,对所述预训练模型进行训练。
例如,文本数据样本为“这个饭菜中有蟑螂”,属于该文本数据样本的服务问题大分类结果样本为“异物”,其为正样本。不属于该文本数据样本的服务问题大分类结果样本可以是“变质”、或者“食源性疾病”、或者“过期”、或者“餐品未熟”,等等,其为负样本。
正样本是需要训练后的大分类文本数据特征分析模型能够查询出的属于文本数据样本的服务问题大分类结果样本,负样本是作为正样本的对比样本,使得大分类文本数据特征分析模型识别不属于文本数据样本的服务问题大分类结果。采用正样本和负样本对大分类文本数据特征分析模型进行训练,使得大分类文本数据样本数据特征分析模型对文本数据对应的服务问题大分类结果的分类准确率得到提升。
上述第三步中采用针对服务信息的文本数据样本和针对所述文本数据样本的服务问题小分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得服务问题归因模型。其中,此处提到的“针对所述文本数据样本的服务问题小分类结果样本”是指属于文本数据样本的服务问题小分类结果,其作为正样本。并且,此处属于文本数据样本的服务问题小分类结果是第二步中确定的服务问题大分类结果中的多个候选服务问题小分类结果中的至少一种结果。
此外,还可以根据针对服务信息的文本数据样本和不属于所述文本数据样本的服务 问题小分类结果样本作为第二负样本对,对所述大分类文本数据特征分析模型进行训练。
例如,文本数据样本为“这个饭菜中有蟑螂”,属于该文本数据样本的服务问题大分类结果样本为“异物”,属于该文本数据样本的服务问题小分类结果样本为“蟑螂”。不属于该文本数据样本的服务问题小分类结果样本可以是“尖锐异物”、或者“老鼠”,等等,其为负样本。
采用正样本和负样本对所述大分类文本数据特征分析模型进行训练,使得获得的服务问题归因模型对文本数据对应的服务问题分类结果的分类准确率得到提升。
本申请实施例提供一种服务问题归因方法,包括:获得针对服务信息的待分析文本数据;将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
在本申请的一个实施例中,服务问题归因模型根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,分析获得针对待分析文本数据的服务问题分类结果。上述方法中提供的服务问题归因模型,将待分析文本数据和与待分析文本数据相关的服务知识图谱一起作为输入信息,服务知识图谱能获取待分析文本数据中的服务信息的属性信息,从而提升对待分析文本数据的语言表达逻辑的理解度。基于服务知识图谱有助于提升服务问题归因模型确定待分析文本数据的语言表达逻辑的准确度,服务问题归因模型对待分析文本数据匹配服务问题分类结果的准确度也得到提升。
第一实施例
本申请第一实施例中提供一种服务问题归因方法,具体流程如图8所示,其为本申请第一实施例中提供的一种服务问题归因方法的流程图。图8所示的服务问题归因方法,包括:步骤S801至步骤S802。
如图8所示,在步骤S801中,获得针对服务信息的待分析文本数据。
本步骤用于获得针对服务信息的待分析文本数据。服务端获得待分析文本数据后,是后续步骤中服务问题归因模型确定待分析文本数据对应的服务问题分类结果的数据基础。
其中,所述服务信息可以为商户向用户提供的针对食品服务的食品服务信息;所述针对服务信息的分析文本数据可以为用户针对所述食品服务信息的评价信息;所述获得针对服务信息的待分析文本数据,包括:获得用户端发送的针对食品服务信息的待分析文本数据。
例如,服务信息以商户为用户提供的餐品为例,针对服务信息的待分析文本数据为用户对其食用餐品的评价信息。
如图8所示,在步骤S802中,将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果。
本步骤用于根据服务问题归因模型,获取所述待分析文本数据对应的服务问题分类结果。以下描述所述服务问题归因模型根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,获得针对待分析文本数据的服务问题分类结果的第一种方法。
其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。服务知识谱图有助于理解待分析文本数据的语言表达逻辑,提升分析待分析文本数据的语言表达逻辑的准确度。因此,将待分析文本数据输入服务问题归因模型中,服务问题归因模型根据待分析文本数据,以及与待分析文本数据相关的服务知识图谱,确定待分析文本数据对应的服务问题分类结果。
需要说明的是,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:
步骤1-1:将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第一嵌入向量,所述第一嵌入向量包括待分析文本数据嵌入向量和与所述待分析文本数据相关的服务知识图谱嵌入向量;
步骤1-2:根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
其中,在步骤1-1中,所述将所述待分析文本数据输入到服务问题归因模型中,获 得针对所述待分析文本数据的第一嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量;根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量;根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第一嵌入向量。
需要说明的是,服务问题归因模型获得待分析文本数据,获取与待分析文本数据相关的服务知识图谱,将待分析文本数据和服务知识图谱一起作为输入信息,分析确定待分析文本数据的服务问题分类结果。其中,服务问题归因模型将待分析文本数据的数据形式转换为待分析文本数据嵌入向量,将服务知识图谱的数据形式转换为服务知识图谱嵌入向量。根据待分析文本数据嵌入向量和与待分析文本数据相关的服务知识图谱嵌入向量,生成针对待分析文本数据的第一嵌入向量。
其中,获取与待分析文本相关的服务知识图谱可以是从服务问题归因模型预先存储的服务知识图谱库中查询到与待分析文本数据相关的服务知识图谱,也可以是从其他共享平台获取与待分析文本数据相关的服务知识图谱。
其中,所述将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;将所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,以及每个文本单元在所述待分析文本数据中对应的位置嵌入向量进行处理,获得所述待分析文本数据嵌入向量。
其中,所述根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据中的服务信息嵌入向量;根据所述待分析文本数据中的服务信息嵌入向量,查询针对所述服务信息的服务知识图谱;获取针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量;将针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量进行处理,获得针对所述服务信息的服务知识图谱嵌入向量。
在步骤1-2中,根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,至少可以通过如下三种方式获得,以下分别论述。
(一)、所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,可以是按照如下第一种方式获得:
根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
其中,所述根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果,包括:获取所述服务问题归因模型的服务问题分类列表中多个候选服务问题分类结果分别对应的特征信息;将所述待分析文本数据的特征信息与所述多个候选服务问题分类结果分别对应的特征信息进行比较,将包含所述待分析文本数据的特征信息的候选服务问题分类结果确定为与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
例如,待分析文本数据为“这个水果饮品中的水果上面出现霉点了”,根据第一嵌入向量确定待分析文本数据的特征信息为“水果和霉点”。
第一嵌入向量包括待分析文本数据嵌入向量和服务知识图谱嵌入向量,此处待分析文本数据嵌入向量包括文本嵌入向量,位置嵌入向量以及段落嵌入向量。第一嵌入向量是将待分析文本嵌入向量和服务知识图谱嵌入向量结合获得的嵌入向量。因此,根据第一嵌入向量分析待分析文本数据的特征信息,需要结合待分析文本数据的语义和待分析文本数据对应的服务知识图谱,确定待分析文本数据的综合语义。
此处,待分析文本数据“这个水果饮品中的水果上面出现霉点了”对应的服务知识图谱为“水果属于新鲜果蔬,新鲜果蔬表面出现霉点属于霉变现象”。因此,根据第一 嵌入向量可知,待分析文本数据的特征信息为“水果和霉点”。
而服务问题分类列表中多个候选服务问题分类结果,如图2所示的“可见发霉变质”和“腐烂”这两个候选服务问题分类结果。其中,“可见发霉变质”的特征信息包括“物体表面出现霉变特征”,“腐烂”的特征信息包括“物体内部出现了大量病变”。将待分析文本数据的特征信息与上述两个候选服务问题分类结果的特征信息进行比较,得到待分析文本数据“这个水果饮品中的水果上面出现霉点了”对应的服务问题分类结果为可见发霉变质。
(二)、所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,可以是按照如下第二种方式获得:
根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的至少一种服务问题分类结果;将所述至少一种服务问题分类结果作为所述待分析文本数据对应的服务问题分类结果。
(三)、所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,可以是按照如下第三种方式获得:
根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题大分类结果,作为所述待分析文本数据的目标服务问题大分类结果;在确定所述待分析文本数据的目标服务问题大分类结果的基础上,在所述目标服务问题大分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题小分类结果;根据所述目标服务问题大分类结果和所述目标服务问题小分类结果,生成所述待分析文本数据对应的服务问题分类结果。
其中,所述服务问题归因模型包括大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题大分类结果的文本数据的特征信息;所述根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题大分类结果,作为所述待分析文本数据的目标服务问题大分类结果,包括:根据所述大分类文本数据特征分析模型,获取所述待分析文本数据中用于分析服务问题大分类结果的文本数据的第一特征信息;获取所述服务问题大分类列表中每个候选服务问题大分类结果对应的文本数据的第一候选特征信息;将所述待分析文本数据中的第一特征信息与所述每个候选服务问题大分类结果对应的第一候选特征信息进行比较,获得所述待分析文本数据对应的目标服务问题大分类结果。
其中,所述在确定所述待分析文本数据的目标服务问题大分类结果的基础上,在所述目标服务问题大分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题小分类结果,包括:获取所述待分析文本数据中用于分析服务问题小分类结果的文本数据的第二特征信息;获取所述服务问题小分类列表中每个候选服务问题小分类结果对应的文本数据的第二候选特征信息;将所述待分析文本数据中的第二特征信息与所述每个候选服务问题小分类结果对应的第二候选特征信息进行比较,获得所述待分析文本数据对应的目标服务问题小分类结果。
以上即为所述服务问题归因模型根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,获得针对待分析文本数据的服务问题分类结果的第一种方法。根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,改善分析待分析文本数据的语言表达逻辑的准确度,在此基础上,提升分析待分析文本数据对应的服务问题分类结果的准确度。
上述第一种方式是基于服务知识图谱,提升分析待分析文本数据的语言表达逻辑的准确度。此外,所述服务问题归因模型获得待分析文本数据对应的服务问题分类结果还包括第二种方式:
所述服务问题归因模型具体用于根据所述待分析文本数据,所述待分析文本数据对应的拼音信息,以及与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
第二种方式中,分析待分析文本数据对应的服务问题分类结果,不仅结合待分析文本数据的服务知识图谱,还结合待分析文本数据的拼音信息,改善待分析文本数据的错别字等问题,也提升了分析待分析文本数据的语言表达逻辑的准确度。
其中,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分 析文本数据的服务问题分类结果,包括:
步骤2-1:将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,所述第二嵌入向量包括所述待分析文本数据的拼音嵌入向量和与所述待分析文本数据相关的服务知识图谱的拼音嵌入向量;
步骤2-2:根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
其中,在步骤2-1中,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析文本数据中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括所述服务知识图谱中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第二嵌入向量。
需要说明的是,服务问题归因模型将待分析文本数据的数据形式转换为嵌入向量的形式,除了获取待分析文本数据的文本嵌入向量,位置嵌入向量,以及段落嵌入向量,还会获取文本单元的拼音嵌入向量。如图1中的待分析文本数据“这个饭菜中有蟑螂”,其包含了每个字的文本嵌入向量,段落嵌入向量,位置嵌入向量以及拼音嵌入向量。文本单元的拼音嵌入向量的加入,能够改善识别待分析文本数据的文字时出现错别字的几率。然后,将待分析文本数据嵌入向量和与待分析文本数据相关的服务知识图谱嵌入向量结合,生成针对待分析文本数据的第二嵌入向量。该过程中,拼音嵌入向量和服务知识图谱结合,不仅提升待分析文本数据中的文字识别准确率,也提升了分析待分析文本数据的语言逻辑表达的准确率。
其中,步骤201获取的第二嵌入向量是由待分析文本数据嵌入向量和服务知识图谱嵌入向量进行向量处理后获得的,此步骤中的待分析文本数据嵌入向量包括待分析文本数据的拼音嵌入向量。
所述将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,包括:将所述待分析文本数据输入所述服务问题归因模型中,获得所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;将所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,以及每个文本单元在所述待分析文本数据中对应的位置嵌入向量进行处理,获得所述待分析文本数据嵌入向量。
所述根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据中的服务信息嵌入向量;根据所述待分析文本数据中的服务信息嵌入向量,查询针对所述服务信息的服务知识图谱;获取针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量;将针对所述服务信息的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中对应的段落嵌入向量,每个文本单元在所述服务知识图谱中对应的位置嵌入向量进行处理,获得针对所述服务信息的服务知识图谱嵌入向量。
以上即为服务问题归因模型通过第二种方式获得待分析文本数据的服务问题分类结果。具体的,服务问题归因模型将待分析文本数据转换成嵌入向量时,在原本获得文本嵌入向量,段落嵌入向量,位置嵌入向量的基础上,结合每个文本单元对应的拼音嵌入向量,减少待分析文本数据中的错别字问题。同时,结合待分析文本数据对应的服务知识图谱,提升分析待分析文本数据的语言表达逻辑的准确度。在确定待分析文本数据的语言表达逻辑的基础上,获取待分析文本数据对应的服务问题大分类结果。在确定的服务问题大分类结果的多个候选服务问题小分类列表中,继续匹配待分析文本数据对应的服务问题小分类结果,根据服务问题大分类结果和服务问题小分类结果,生成待分析文本数据对应的服务问题分类结果。
本申请实施例提供一种服务问题归因方法,包括:获得针对服务信息的待分析文本数据;将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据和 与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
上述方法,服务问题归因模型根据待分析文本数据,和与待分析文本数据相关的服务知识图谱,分析获得针对待分析文本数据的服务问题分类结果。上述方法中提供的服务问题归因模型,将待分析文本数据和与待分析文本数据相关的服务知识图谱一起作为输入信息,服务知识图谱能获取待分析文本数据中的服务信息的属性信息,从而提升对待分析文本数据的语言表达逻辑的理解度。基于服务知识图谱有助于提升服务问题归因模型确定待分析文本数据的语言表达逻辑的准确度,服务问题归因模型对待分析文本数据匹配服务问题分类结果的准确度也得到提升。
第二实施例
与本申请提供的服务问题归因方法的应用场景对应的实施例以及第一实施例提供的服务问题归因方法相对应的,本申请第二实施例还提供了一种服务问题归因装置。由于装置实施例基本相似于应用场景对应的实施例以及第一实施例,所以描述得比较简单,相关之处参见应用场景对应的实施例以及第一实施例的部分说明即可。下述描述的装置实施例仅仅是示意性的。
请参考图9,其为本申请第二实施例中提供的一种服务问题归因装置的示意图。本申请第二实施例中提供的一种服务问题归因装置,包括:
第一获得单元901,用于获得针对服务信息的待分析文本数据。
第二获得单元902,用于将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
第三实施例
与本申请提供的服务问题归因方法的应用场景对应的实施例以及第一实施例提供的服务问题归因方法相对应的,本申请第三实施例还提供了另一种服务问题归因方法。
请参照图10,其为本申请第三实施例中提供的另一种服务问题归因方法的示意图。图10所示的服务问题归因方法,包括:步骤S1001至步骤S1002。
如图10所示,在步骤S1001中,获得针对服务信息的待分析文本数据。
如图10所示,在步骤S1002中,将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果。
其中,所述服务问题归因模型用于根据所述待分析文本数据以及所述待分析文本数据对应的拼音信息,获得针对所述待分析文本数据的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:将所述待分析文本数据输入到服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得待分析文本数据嵌入向量,包括:将所述待分析文本数据输入到服务问题归因模型中,对所述待分析文本数据进行向量化处理,获得所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;对所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量进行处理,获得所述待分析文本数据嵌入向量。
可选的,所述根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
可选的,所述根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述待分析文本数据嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题大分类结果,作为所述待分析文本数据的目标服务问题大分类结果;在确定所述待分析文本数据的目标服务问题大分类结果的基础上,在所述目标服务问题大分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题小分类结果;根据所述目标服务问题大分类结果和所述目标服务问题小分类结果,生成所述待分析文本数据对应的服务问题分类结果。
可选的,所述服务问题归因模型具体用于根据所述待分析文本数据,所述待分析文本数据对应的拼音信息,以及与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,所述第二嵌入向量包括待分析文本数据嵌入向量和与所述待分析文本数据相关的服务知识图谱嵌入向量;根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
可选的,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,包括:将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析文本数据中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括所述服务知识图谱中每个文本单元对应的拼音嵌入向量;根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第二嵌入向量。
可选的,所述根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:根据所述第二嵌入向量,获取所述待分析文本数据的特征信息;根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题大分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题大分类结果,作为所述待分析文本数据的目标服务问题大分类结果;在确定所述待分析文本数据的目标服务问题大分类结果的基础上,在所述目标服务问题大分类结果的服务问题小分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题小分类结果;根据所述目标服务问题大分类结果和所述目标服务问题小分类结果,生成所述待分析文本数据对应的服务问题分类结果。
第四实施例
与本申请提供的服务问题归因方法的应用场景对应的实施例以及第三实施例提供的服务问题归因方法相对应的,本申请第四实施例还提供了另一种服务问题归因装置。由于装置实施例基本相似于应用场景对应的实施例以及第三实施例,所以描述得比较简单,相关之处参见应用场景对应的实施例以及第三实施例的部分说明即可。下述描述的装置实施例仅仅是示意性的。
请参考图11,其为本申请第四实施例中提供的另一种服务问题归因装置的示意图。本申请第四实施例中提供的另一种服务问题归因装置,包括:
第三获得单元1101,用于获得针对服务信息的待分析文本数据。
第四获得单元1102,用于将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果。其中,所述服务问题归因模型用于根据所述待分析文本数据以及所述待分析文本数据对应的拼音信息,获得针对所述待分析文本数据的服务问题分类结果。
第五实施例
与本申请提供的服务问题归因方法的应用场景对应的实施例以及第一实施例提供的服务问题归因方法相对应的,本申请第五实施例还提供了一种服务问题归因模型的训练方法。
请参照图12,其为本申请第五实施例中提供的一种服务问题归因模型的训练方法的示意图。图12所示的服务问题归因方法,包括:步骤S1201至步骤S1203。
如图12所示,在步骤S1201中,获得用于分析文本数据的特征信息的预训练模型。
如图12所示,在步骤S1202中,根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题大分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题大分类结果的文本数据的特征信息。
如图12所示,在步骤S1203中,根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题小分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
可选的,所述预训练模型通过如下方式分析文本数据的特征信息:根据所述文本数据获得文本数据嵌入向量,所述文本数据嵌入向量包括所述文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述文本数据中的段落嵌入向量,以及每个文本单元在所述文本数据中的位置嵌入向量;根据所述文本数据嵌入向量,分析所述文本数据的特征信息。
可选的,还包括:根据所述文本数据嵌入向量获取与所述文本数据相关联的服务知识图谱嵌入向量,所述服务知识图谱嵌入向量包括与所述文本数据相关联的服务知识图谱中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述服务知识图谱中的段落嵌入向量,以及每个文本单元在所述服务知识图谱中的位置嵌入向量;所述根据所述文本数据嵌入向量,分析所述文本数据的特征信息,包括:根据所述文本数据嵌入向量,以及与所述文本数据相关联的服务知识图谱嵌入向量,分析所述文本数据的特征信息。
可选的,所述根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题大分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,包括:将所述针对服务信息的文本数据样本输入所述预训练模型中,获得所述预训练模型输出的针对所述服务信息的文本数据样本的第一服务问题大分类结果;根据所述第一服务问题大分类结果和所述针对所述文本数据样本的服务问题大分类结果样本之间的相似程度,对所述预训练模型进行服务问题大分类参数的调整,获得所述大分类文本数据特征分析模型。
可选的,所述将所述针对服务信息的文本数据样本输入所述预训练模型中,获得所述预训练模型输出的针对所述服务信息的文本数据样本的第一服务问题大分类结果,包括:将所述针对服务信息的文本数据样本输入所述预训练模型中,获得所述文本数据样本的特征信息;根据所述文本数据样本的特征信息,获得所述预训练模型输出的针对所述服务信息的文本数据样本的第一服务问题大分类结果。
可选的,所述根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题小分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型,包括:将针对所述服务信息的文本数据输入所述大分类文本数据特征分析模型中,获得所述大分类文本数据特征分析模型输出的针对所述文本数据样本的第二服务问题小分类结果;根据所述第二服务问题小分类结果与所述服务问题小分类结果样本之间的相似程度,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
可选的,还包括:将针对服务信息的文本数据样本和不属于所述文本数据的服务问题大分类结果样本作为第一负样本对,对所述预训练模型进行调整,获得用于分析对应于服务问题大分类结果的文本数据的特征信息的大分类文本数据特征分析模型;将针对服务信息的文本数据样本和不属于所述文本数据的服务问题小分类结果样本作为第二负样本对,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
第六实施例
与本申请提供的服务问题归因方法的应用场景对应的实施例以及第五实施例提供的服务问题归因模型的训练方法相对应的,本申请第六实施例还提供了一种服务问题归因模型的训练装置。由于装置实施例基本相似于应用场景对应的实施例以及第五实施例,所以描述得比较简单,相关之处参见应用场景对应的实施例以及第五实施例的部分说明即可。下述描述的装置实施例仅仅是示意性的。
请参考图13,其为本申请第六实施例中提供的一种服务问题归因模型的训练装置的示意图。本申请第六实施例中提供的一种服务问题归因模型的训练装置,包括:
预训练模型获得单元1301,用于获得用于分析文本数据的特征信息的预训练模型。
大分类文本数据特征分析模型获得单元1302,用于根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题大分类结果样本,对所述预训练模型进行调整,获得大分类文本数据特征分析模型,所述大分类文本数据特征分析模型用于分析对应于服务问题大分类结果的文本数据的特征信息。
服务问题归因模型获得单元1303,用于根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题小分类结果样本,对所述大分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
第七实施例
与本申请提供的上述方法实施例相对应的,本申请第七实施例还提供了一种电子设备。由于第七实施例基本相似于本申请提供的上述方法实施例,所以描述得比较简单,相关之处参见本申请提供的上述方法实施例部分的说明即可。下面描述的第七实施例仅仅是示意性的。
请参照图14,其为本申请第七实施例中提供的一种电子设备的示意图。该电子设备,包括:至少一个处理器1401,至少一个通信接口1402,至少一个存储器1403和至少一个通信总线1404;可选的,通信接口1402可以为通信模块的接口,如WLAN(Wireless Local Area Network,无线局域网)模块的接口;处理器140l可能是处理器CPU(Central Processing Unit,中央处理器),或者是特定集成电路ASIC(Application Specific Integrated Circuit),或者是被配置成实施本申请实施例的一个或多个集成电路。存储器1403可能包含高速RAM(Random Access Memory,随机存取存储器),也可能还包括非易失性存储器(non-volatile memory),例如至少一个磁盘存储器。其中,存储器1403存储有程序,处理器1401调用存储器1403所存储的程序,以执行本申请实施例提供的上述方法。
第八实施例
与本申请提供的上述方法实施例相对应的,本申请第八实施例还提供了一种计算机存储介质,由于第八实施例基本相似于本申请提供的上述方法实施例,所以描述的比较简单,相关之处参见本申请提供的上述方法实施例部分的说明即可。下面描述的第八实施例仅仅是示意性的。所述计算机存储介质存储有计算机程序,所述程序被执行时实现上述方法实施例中提供的方法。需要说明的是,本申请第八实施例提供的存储介质的详细描述,可以参考对本申请提供的上述方法实施例的相关描述,这里不再赘述。
本申请虽然以较佳实施例公开如上,但其并不是用来限定本申请,任何本领域技术人员在不脱离本申请的精神和范围内,都可以做出可能的变动和修改,因此本申请的保护范围应当以本申请权利要求所界定的范围为准。
在一个典型的配置中,计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。内存可能包括计算机可读介质中的非永久性存储器,随机存取存储器(RAM)和/或非易失性内存等形式,如只读存储器(Read-Only Memory,ROM)或闪存(f;ash RAM)。内存是计算机可读介质的示例。
1、计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括,但不限于相变内存(Phase-change RAM,PRAM)、静态随机存取存储器(Static RAM,SRAM)、动态随机存取存储器(Dynamic RAM,DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(Electrically Erasable Programmable ROM,EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带,磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质,可用于存储可以被计算设备访问的信息。按照本文中的界定,计算机可读介质不包括非暂存电脑可读媒体(Transitory Media),如调制的数据信号和载波。

Claims (10)

  1. 一种服务问题归因方法,其特征在于,包括:
    获得针对服务信息的待分析文本数据;
    将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;
    其中,所述服务问题归因模型用于根据所述待分析文本数据和与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
  2. 根据权利要求1所述的方法,其特征在于,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:
    将所述待分析文本数据输入到所述服务问题归因模型中,获得针对所述待分析文本数据的第一嵌入向量,所述第一嵌入向量包括待分析文本数据嵌入向量和与所述待分析文本数据相关的服务知识图谱嵌入向量;
    根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
  3. 根据权利要求2所述的方法,其特征在于,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第一嵌入向量,包括:
    将所述待分析文本数据输入到所述服务问题归因模型中,获得所述待分析文本数据嵌入向量;
    根据所述待分析文本数据嵌入向量,查询与所述待分析文本数据嵌入向量相关联的所述服务知识图谱嵌入向量;
    根据所述待分析文本数据嵌入向量和所述服务知识图谱嵌入向量,获得针对所述待分析文本数据的第一嵌入向量。
  4. 根据权利要求2所述的方法,其特征在于,所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:
    根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;
    根据所述待分析文本数据的特征信息,在所述服务问题归因模型的服务问题分类列表中查询与所述待分析文本数据的特征信息匹配的目标服务问题分类结果,作为所述待分析文本数据对应的服务问题分类结果。
  5. 根据权利要求2所述的方法,其特征在于,所述根据所述第一嵌入向量,获取所述待分析文本数据对应的服务问题分类结果,包括:
    根据所述第一嵌入向量,获取所述待分析文本数据的特征信息;
    根据所述待分析文本数据的特征信息,在所述服务问题归因模型中的服务问题第一分类列表中查询与所述待分析文本数据的特征信息相关联的服务问题第一分类结果,作为所述待分析文本数据的目标服务问题第一分类结果;
    在确定所述待分析文本数据的目标服务问题第一分类结果的基础上,在所述目标服务问题第一分类结果的服务问题第二分类列表中查询与所述待分析文本数据的特征信息相关联的目标服务问题第二分类结果;
    根据所述目标服务问题第一分类结果和所述目标服务问题第二分类结果,生成所述待分析文本数据对应的服务问题分类结果。
  6. 根据权利要求1所述的方法,其特征在于,所述服务问题归因模型用于根据所述待分析文本数据,所述待分析文本数据对应的拼音信息,以及与所述待分析文本数据相关的服务知识图谱,获得针对所述待分析文本数据的服务问题分类结果。
  7. 根据权利要求6所述的方法,其特征在于,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:
    将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的第二嵌入向量,所述第二嵌入向量包括所述待分析文本数据的拼音嵌入向量和与所述待分析文本数据相关的服务知识图谱的拼音嵌入向量;
    根据所述第二嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
  8. 一种服务问题归因方法,其特征在于,包括:
    获得针对服务信息的待分析文本数据;
    将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果;
    其中,所述服务问题归因模型用于根据所述待分析文本数据以及所述待分析文本数据对应的拼音信息,获得针对所述待分析文本数据的服务问题分类结果。
  9. 根据权利要求8所述的方法,其特征在于,所述将所述待分析文本数据输入到服务问题归因模型中,获得针对所述待分析文本数据的服务问题分类结果,包括:
    将所述待分析文本数据输入到所述服务问题归因模型中,获得待分析文本数据嵌入向量,所述待分析文本数据嵌入向量包括所述待分析文本数据中每个文本单元对应的文本嵌入向量,每个文本单元对应的拼音嵌入向量,每个文本单元在所述待分析文本数据中对应的段落嵌入向量,每个文本单元在所述待分析文本数据中对应的位置嵌入向量;
    根据所述待分析文本数据嵌入向量,获取所述待分析文本数据对应的服务问题分类结果。
  10. 一种服务问题归因模型的训练方法,其特征在于,包括:
    获得用于分析文本数据的特征信息的预训练模型;
    根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第一分类结果样本,对所述预训练模型进行调整,获得第一分类文本数据特征分析模型,所述第一分类文本数据特征分析模型用于分析对应于服务问题第一分类结果的文本数据的特征信息;
    根据针对服务信息的文本数据样本和针对所述文本数据样本的服务问题第二分类结果样本,对所述第一分类文本数据特征分析模型进行调整,获得用于分析针对待分析文本数据的服务问题分类结果的服务问题归因模型。
PCT/CN2023/103688 2022-06-29 2023-06-29 一种服务问题归因方法及装置 WO2024002216A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210763456.8A CN115062605A (zh) 2022-06-29 2022-06-29 一种服务问题归因方法及装置
CN202210763456.8 2022-06-29

Publications (1)

Publication Number Publication Date
WO2024002216A1 true WO2024002216A1 (zh) 2024-01-04

Family

ID=83204508

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/103688 WO2024002216A1 (zh) 2022-06-29 2023-06-29 一种服务问题归因方法及装置

Country Status (2)

Country Link
CN (1) CN115062605A (zh)
WO (1) WO2024002216A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115062605A (zh) * 2022-06-29 2022-09-16 拉扎斯网络科技(上海)有限公司 一种服务问题归因方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111177416A (zh) * 2020-04-13 2020-05-19 傲林科技有限公司 事件根因分析模型构建方法、事件根因分析方法及装置
WO2021107448A1 (ko) * 2019-11-25 2021-06-03 주식회사 데이터마케팅코리아 효율적 문서 분류 처리를 지원하는 지식 그래프 기반 마케팅 정보 분석 서비스 제공 방법 및 그 장치
CN114429284A (zh) * 2021-12-31 2022-05-03 拉扎斯网络科技(上海)有限公司 线上控制食品风险的方法、服务器、终端及电子设备
CN115062605A (zh) * 2022-06-29 2022-09-16 拉扎斯网络科技(上海)有限公司 一种服务问题归因方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021107448A1 (ko) * 2019-11-25 2021-06-03 주식회사 데이터마케팅코리아 효율적 문서 분류 처리를 지원하는 지식 그래프 기반 마케팅 정보 분석 서비스 제공 방법 및 그 장치
CN111177416A (zh) * 2020-04-13 2020-05-19 傲林科技有限公司 事件根因分析模型构建方法、事件根因分析方法及装置
CN114429284A (zh) * 2021-12-31 2022-05-03 拉扎斯网络科技(上海)有限公司 线上控制食品风险的方法、服务器、终端及电子设备
CN115062605A (zh) * 2022-06-29 2022-09-16 拉扎斯网络科技(上海)有限公司 一种服务问题归因方法及装置

Also Published As

Publication number Publication date
CN115062605A (zh) 2022-09-16

Similar Documents

Publication Publication Date Title
Tao et al. Utilization of text mining as a big data analysis tool for food science and nutrition
Nguyen et al. Building a national neighborhood dataset from geotagged Twitter data for indicators of happiness, diet, and physical activity
Singh et al. Social media data analytics to improve supply chain management in food industries
JP6182279B2 (ja) データ分析システム、データ分析方法、データ分析プログラム、および、記録媒体
US20150286710A1 (en) Contextualized sentiment text analysis vocabulary generation
US10592540B2 (en) Generating elements of answer-seeking queries and elements of answers
Ray et al. Exploring the drivers of customers’ brand attitudes of online travel agency services: A text-mining based approach
CN116821308B (zh) 生成方法、模型的训练方法、设备及存储介质
Merler et al. Snap, Eat, RepEat: A food recognition engine for dietary logging
Hossain et al. Customer sentiment analysis and prediction of halal restaurants using machine learning approaches
WO2024002216A1 (zh) 一种服务问题归因方法及装置
Ayyub et al. Drivers of organic food purchase intention in a developing country: the mediating role of trust
Donadello et al. Ontology-driven food category classification in images
Roither et al. The chef’s choice: system for allergen and style classification in recipes
Zuo et al. A food safety prescreening method with domain-specific information using online reviews
JP5933863B1 (ja) データ分析システム、制御方法、制御プログラム、および記録媒体
Hafez et al. Multi-criteria recommendation systems to foster online grocery
JP2017201543A (ja) データ分析システム、データ分析方法、データ分析プログラム、および、記録媒体
AU2020202979A1 (en) System for improved remote processing and interaction with artificial survey administrator
Qiu et al. A deep matching model for detecting reviews mismatched with products in e-commerce
Shi et al. Attention-based ingredient phrase parser
Bhattacharjee et al. What drives consumer choices? Mining aspects and opinions on large scale review data using distributed representation of words
Wang et al. Understanding public perceptions of measles from twitter using multi-task convolutional neural networks
Lee et al. Deriving topic-related and interaction features to predict top attractive reviews for a specific business entity
Verma et al. A survey on sentiment analysis techniques for twitter

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23830385

Country of ref document: EP

Kind code of ref document: A1