CN117609457A

CN117609457A - Information processing method and device, storage medium and electronic equipment

Info

Publication number: CN117609457A
Application number: CN202311589776.7A
Authority: CN
Inventors: 熊步先
Original assignee: Industrial and Commercial Bank of China Ltd ICBC
Current assignee: Industrial and Commercial Bank of China Ltd ICBC
Priority date: 2023-11-24
Filing date: 2023-11-24
Publication date: 2024-02-27

Abstract

The application discloses an information processing method and device, a storage medium and electronic equipment, and relates to the technical field of artificial intelligence, the field of financial science and technology or other related fields. The method comprises the following steps: acquiring N text messages; determining category information corresponding to the N text information according to the N text information and the target classification model; determining a target instruction according to category information corresponding to the N pieces of text information, inputting the N pieces of text information and the target instruction into an information extraction model for processing, and obtaining target information in the N pieces of text information; and determining answer information corresponding to the question information based on the target information in the N text information and the information in the database, and returning the answer information to the first object. According to the method and the device for answering the financial related questions, the problem that in the related technology, the questions related to finance are answered in a mode of manually inquiring related files through customer service is solved, and the problem that the efficiency of answering the questions is low is solved.

Description

Information processing method and device, storage medium and electronic equipment

Technical Field

The present application relates to the field of artificial intelligence, financial technology, or other related fields, and in particular, to an information processing method and apparatus, a storage medium, and an electronic device.

Background

Currently, for telephone service personnel, various questions are presented during service, and in many cases, the questions can be accurately answered only after related files are queried. However, the manual inquiry of related files by means of customer service is laborious and laborious. Therefore, with the development of technology, knowledge-based follower methods based on artificial intelligence technology in related technology are increasingly widely used. The method utilizes a traditional deep learning model, can automatically match according to dialogue, and gives out information of related problems. However, conventional deep learning models in the related art, such as LSTM (Long Short-Term Memory Network, long Short term memory network), BERT (Bidirectional Encoder Representations from Transformers, pre-training language model), fastrext (a machine learning model for natural language processing), rely on training corpus, and when the corpus coverage is not wide enough, the model effect is poor, so that it is difficult for the conventional deep learning model to answer the customer questions faster and better.

Aiming at the problem that the problem related to finance, which is presented by a client and is related to the finance, is answered by manually inquiring related files through customer service in the related technology, the problem that the efficiency of answering the problem is lower is solved, and no effective solution is presented at present.

Disclosure of Invention

The main objective of the present application is to provide an information processing method and apparatus, a storage medium, and an electronic device, so as to solve the problem in the related art that the problem related to finance is solved by manually querying the related file through customer service, resulting in lower efficiency of answering the problem.

In order to achieve the above object, according to one aspect of the present application, there is provided an information processing method. The method comprises the following steps: acquiring N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object which provides financial services for the first object in a financial institution, and N is a positive integer greater than 1; determining category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model; determining a target instruction according to category information corresponding to the N pieces of text information, and inputting the N pieces of text information and the target instruction into an information extraction model for processing to obtain target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information; and determining answer information corresponding to the question information based on target information in the N pieces of text information and information in a database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information.

Further, determining answer information corresponding to the question information based on the target information in the N text information and information in the database includes: inputting target information in the N pieces of text information into a search engine to obtain S pieces of information in the database, wherein the search engine is used for searching information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than the similarity between each piece of information in the T pieces of information and the target information, the T pieces of information are information except the S pieces of information in the database, S is a positive integer greater than 1, and T is a positive integer; and determining answer information corresponding to the question information based on the S pieces of information.

Further, inputting the N text information and the target instruction into an information extraction model for processing, and obtaining target information in the N text information includes: extracting information from the N pieces of text information according to the target instruction through the information extraction model to obtain information extracted from the N pieces of text information; and taking the information extracted from the N pieces of text information as the target information in the N pieces of text information.

Further, determining category information corresponding to the N text information according to the N text information and the target classification model includes: performing word segmentation processing on the N text messages to obtain N segmented text messages; inputting the text information subjected to the N word segmentation into the target classification model for classification processing to obtain category information corresponding to the N text information.

Further, the object classification model is obtained by: obtaining M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1; labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample; determining training data according to the M segmented text samples and category information corresponding to each segmented text sample; and learning and training the original classification model by utilizing the training data to obtain the target classification model.

Further, performing learning training on the original classification model by using the training data, and obtaining the target classification model includes: learning and training the original classification model by utilizing the training data to obtain a trained classification model; judging whether the trained classification model meets a preset stopping condition, wherein the stopping condition is a condition for stopping training the original classification model; if the trained classification model does not meet the stopping condition, continuing to learn and train the original classification model; and if the trained classification model meets the stopping condition, evaluating the trained classification model to obtain an evaluation result, and obtaining the target classification model based on the evaluation result.

Further, deriving the target classification model based on the evaluation result includes: judging whether the evaluation result accords with an expected result or not; if the evaluation result does not accord with the expected result, adjusting the target data information to obtain adjusted target data information, wherein the target data information at least comprises: the training data and the parameter information of the original classification model; continuing to learn and train the original classification model based on the adjusted target data information; and if the evaluation result accords with the expected result, taking the trained classification model as the target classification model.

Further, acquiring the N pieces of text information includes: acquiring dialogue voice information of dialogue between the first object and the second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object; performing transcription processing on the dialogue voice information to obtain N original text information; preprocessing the N original text messages to obtain the N text messages, wherein the preprocessing is at least one of the following steps: filtering stop words, removing target characters and replacing target words.

In order to achieve the above object, according to another aspect of the present application, there is provided an information processing apparatus. The device comprises: the first obtaining unit is used for obtaining N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object which provides financial services for the first object in a financial institution, and N is a positive integer greater than 1; the first determining unit is used for determining category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model; the first processing unit is used for determining a target instruction according to category information corresponding to the N pieces of text information, inputting the N pieces of text information and the target instruction into an information extraction model for processing, and obtaining target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information; and the second processing unit is used for determining answer information corresponding to the question information based on the target information in the N pieces of text information and information in a database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information.

Further, the second processing unit includes: the first input module is used for inputting target information in the N pieces of text information into a search engine to obtain S pieces of information in the database, wherein the search engine is used for searching information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than the similarity between each piece of information in the T pieces of information and the target information, the T pieces of information are information except the S pieces of information in the database, S is a positive integer greater than 1, and T is a positive integer; and the first determining module is used for determining the answer information corresponding to the question information based on the S pieces of information.

Further, the first processing unit includes: the first extraction module is used for extracting information from the N pieces of text information according to the target instruction through the information extraction model to obtain information extracted from the N pieces of text information; and the second determining module is used for taking information extracted from the N pieces of text information as the target information in the N pieces of text information.

Further, the first determination unit includes: the first processing module is used for carrying out word segmentation processing on the N text messages to obtain N segmented text messages; and the second input module is used for inputting the text information subjected to the N word segmentation into the target classification model for classification processing to obtain category information corresponding to the N text information.

Further, the object classification model is obtained by: the second acquisition unit is used for acquiring M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1; the first labeling unit is used for labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample; the second determining unit is used for determining training data according to the M segmented text samples and category information corresponding to each segmented text sample; and the first training unit is used for learning and training the original classification model by utilizing the training data to obtain the target classification model.

Further, the first training unit includes: the first training module is used for learning and training the original classification model by utilizing the training data to obtain a trained classification model; the first judging module is used for judging whether the trained classification model meets a preset stopping condition, wherein the stopping condition is a condition for stopping training the original classification model; the second training module is used for continuing to learn and train the original classification model if the trained classification model does not meet the stopping condition; and the second processing module is used for evaluating the trained classification model to obtain an evaluation result if the trained classification model meets the stopping condition, and obtaining the target classification model based on the evaluation result.

Further, the second processing module includes: the first judging submodule is used for judging whether the evaluation result accords with an expected result or not; the first adjustment sub-module is configured to adjust the target data information if the evaluation result does not conform to the expected result, so as to obtain adjusted target data information, where the target data information at least includes: the training data and the parameter information of the original classification model; the first training sub-module is used for continuing to learn and train the original classification model based on the adjusted target data information; and the first determination submodule is used for taking the trained classification model as the target classification model if the evaluation result accords with the expected result.

Further, the first acquisition unit includes: the first acquisition module is used for acquiring dialogue voice information of dialogue between the first object and the second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object; the third processing module is used for carrying out transcription processing on the dialogue voice information to obtain N original text information; the fourth processing module is configured to perform preprocessing on the N original text messages to obtain the N text messages, where the preprocessing is at least one of the following: filtering stop words, removing target characters and replacing target words.

In order to achieve the above object, according to another aspect of the present application, there is provided a computer-readable storage medium storing a program, wherein the program performs the information processing method of any one of the above.

In order to achieve the above object, according to another aspect of the present application, there is provided an electronic device including one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement any one of the information processing methods described above.

Through the application, the following steps are adopted: acquiring N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object for providing financial services for the first object in a financial institution, and N is a positive integer greater than 1; determining category information corresponding to the N text information according to the N text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model; determining a target instruction according to category information corresponding to the N pieces of text information, inputting the N pieces of text information and the target instruction into an information extraction model for processing, and obtaining target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information; and determining answer information corresponding to the question information based on the target information in the N pieces of text information and the information in the database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information, so that the problem that in the related technology, the question related to finance is answered by a customer in a mode of manually inquiring related files through customer service, and the problem that the efficiency of answering the question is lower is solved. The method comprises the steps of obtaining a plurality of text messages at least comprising questions presented by clients to customer service, wherein the questions presented by the clients are questions related to finance, determining category information corresponding to the plurality of text messages according to the plurality of text messages and the classification model, determining instructions for extracting key information according to the category information corresponding to the plurality of text messages, inputting the plurality of text messages and the instructions into the information extraction model for processing, obtaining key information in the plurality of text messages, determining answers corresponding to the questions presented by the clients based on the key information in the plurality of text messages and information in a database, and returning the answers to the clients, so that the questions related to finance presented by the clients are answered in a mode of manually inquiring related files by the customer service is not needed, and further, the effect of improving the efficiency of answering the questions of the clients is achieved.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application, illustrate and explain the application and are not to be construed as limiting the application. In the drawings:

FIG. 1 is a flow chart of an information processing method provided according to an embodiment of the present application;

fig. 2 is a schematic diagram of modules corresponding to an information processing method in an embodiment of the present application;

FIG. 3 is a flow chart of training a classification model in an embodiment of the application;

fig. 4 is a schematic diagram of an information processing apparatus provided according to an embodiment of the present application;

fig. 5 is a schematic diagram of an electronic device provided according to an embodiment of the present application.

Detailed Description

It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings in conjunction with embodiments.

In order to make the present application solution better understood by those skilled in the art, the following description will be made in detail and with reference to the accompanying drawings in the embodiments of the present application, it is apparent that the described embodiments are only some embodiments of the present application, not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, shall fall within the scope of the present application.

It should be noted that the terms "first," "second," and the like in the description and claims of the present application and the above figures are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate in order to describe the embodiments of the present application described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.

It should be noted that, related information (including, but not limited to, user equipment information, user personal information, etc.) and data (including, but not limited to, data for presentation, analyzed data, etc.) related to the present disclosure are information and data authorized by a user or sufficiently authorized by each party. For example, an interface is provided between the system and the relevant user or institution, before acquiring the relevant information, the system needs to send an acquisition request to the user or institution through the interface, and acquire the relevant information after receiving the consent information fed back by the user or institution.

For convenience of description, the following will describe some terms or terms related to the embodiments of the present application:

BERT (Bidirectional Encoder Representations from Transformers) is a pre-trained language model for natural language processing tasks. The BERT model learns rich language knowledge by self-supervised learning of large-scale unlabeled text during the training phase. It can generate a contextually relevant representation of each input word so that the model can better understand the relationships and semantics between the words. Moreover, by fine tuning the BERT model, model training and prediction may be performed for specific tasks.

The jieba model refers to a Chinese word segmentation tool for segmenting continuous Chinese text into meaningful words. Moreover, the jieba model is a pre-trained model of the tool, and can be used for word segmentation tasks of Chinese text.

The present invention will be described with reference to preferred implementation steps, and fig. 1 is a flowchart of an information processing method according to an embodiment of the present application, as shown in fig. 1, and the method includes the following steps:

step S101, N pieces of text information are acquired, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object in a financial institution providing financial services for the first object, and N is a positive integer greater than 1.

For example, the first object may be a customer of a financial institution, the second object may be a business person of the financial institution, that is, the second object may be a customer service or a telephone service person, the second object may be a robot that provides a telephone service for the customer, such as a customer service robot, etc., the financial institution may be a financial institution such as a bank, etc., and the financial service may be a telephone service provided by the customer to the customer. That is, when a customer consults a service to a customer service in the form of a voice call, the customer service may give information of the service for a problem posed by the customer, and this process may be the financial service described above. The N text messages may be a plurality of sentence messages obtained after the conversation between the customer service and the customer is transcribed, and the transcribed plurality of sentence messages may include question information that the customer proposes to the customer service. For example, a client who speaks a sentence among the plurality of sentences written may be used as the question information.

Step S102, determining category information corresponding to the N text information according to the N text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model.

For example, it is possible to classify the transcribed sentences based on a plurality of sentence information (the above-described N pieces of text information) obtained after transcription of a dialogue between a customer service and a customer, and a classification model (the above-described target classification model), and determine which scene the sentence of the dialogue between the customer service and the customer belongs to. Further, the category information and the scenario may be credit card type (scenario), loan type (scenario), business transaction type (scenario), or the like. That is, if words related to a credit card or the like are included in a dialogue between a customer service and a customer, it can be determined that the dialogue between the customer and the customer service belongs to a credit card scenario; if words related to loans and the like are included in the dialogue between the customer service and the customer, it can be determined that the dialogue between the customer and the customer service belongs to a loan scenario.

Step S103, determining a target instruction according to category information corresponding to the N pieces of text information, and inputting the N pieces of text information and the target instruction into an information extraction model for processing to obtain target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information.

For example, the information extraction model may be a large model, and the large model includes, but is not limited to, GLM (generalized linear model, generalized Linear Models, for solving different kinds of data analysis problems), GPT (generated Pre-Trained Transformer, a Pre-training model, a natural language processing model based on deep learning), and the like. Further, different instructions (target instructions described above) may be input into a large model (information extraction model described above) for different scenes (categories). For example, for a credit card scenario, the instruction (the target instruction described above) may be "extract name, card number, phone number", while for an account blocked scenario, the instruction (the target instruction described above) may be "extract event keywords, each keyword not exceeding 10 words, not exceeding 5 keywords". The above is merely an example, and the instruction (the target instruction described above) may be set by itself. Therefore, when extracting information in a text (the above-described N pieces of text information), information can be extracted using a large model. Specifically, a plurality of sentences (the N pieces of text information) after conversation between the customer service and the customer and instructions (the target instructions) corresponding to the conversation content (the N pieces of text information) may be input into a large model (the information extraction model), and key information (the target information) extracted from the conversation content (the N pieces of text information) may be output.

Step S104, answer information corresponding to the question information is determined based on the target information in the N text information and information in a database, and the answer information is returned to the first object, wherein the database is used for storing information related to the question information.

For example, key information extracted by using a large model (the information extraction model described above) can be matched with information in the database, for example, customer information can be obtained in the database according to name card number, and the current state of the card; based on event keywords such as account freeze, online banking malfunction, etc., a solution to the problem can be searched in the database. The customer information obtained from the database or the solution to the problem searched in the database is then sent to the customer (the first object described above).

It should be noted that the information processing method provided in the embodiment of the present application may be applied to a financial scenario.

Through the steps S101 to S104, a plurality of text information including at least a question posed by a customer to a customer service is obtained, the question posed by the customer is a question related to finance, category information corresponding to the plurality of text information is determined according to the plurality of text information and the classification model, an instruction for extracting key information is determined according to the category information corresponding to the plurality of text information, the plurality of text information and the instruction are input into the information extraction model for processing, the key information in the plurality of text information is obtained, an answer corresponding to the question posed by the customer is determined based on the key information in the plurality of text information and information in the database, and the answer is returned to the customer, so that the question related to finance posed by the customer is answered in a mode of manually querying related files through the customer service is not needed, and the effect of improving the efficiency of answering the customer questions is achieved.

Optionally, in the information processing method provided in the embodiment of the present application, acquiring N pieces of text information includes: the method comprises the steps of obtaining dialogue voice information of dialogue between a first object and a second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object; performing transcription processing on the dialogue voice information to obtain N original text information; preprocessing N original text messages to obtain N text messages, wherein the preprocessing is at least one of the following: filtering stop words, removing target characters and replacing target words.

For example, the embodiment may be a real-time knowledge accompanying method, so that the input voice may be first transcribed in real time, and then the transcribed text may be analyzed. The customer service and the customer speech channels are different, and the customer service and the customer speech are alternately processed as a distinction in the transfer, and each character is transferred into a sentence. Namely, customer service: sense 1, client: sense 2, customer service: sendence 3.

For example, the obtained text data (i.e., the above-mentioned text 1, text 2, and text 3, etc.) may then be cleaned by the data cleaning module, and the cleaning step may have three steps: (1) Stop word filtering, filtering words which have high occurrence times but have no significant meaning in text information, including imaginary words such as word and phrase, adverbs, preposition and the like, and high-frequency words such as' yes, no, and the like, wherein the stop words needing filtering are set by themselves. (2) Digits, english and special characters (corresponding to the target characters) in the text are removed. (3) The special vocabulary (corresponding to the target vocabulary described above) is replaced. When the special vocabulary is transcribed, errors can occur in transcription, for example, an A1 bank can be transcribed into an A2 bank, and substitution processing is carried out on the important special vocabulary, so that the special vocabulary is set automatically.

In summary, through the transcription process and the cleaning process for the dialogue voice between the customer service and the customer, the text information corresponding to the dialogue voice can be obtained rapidly and accurately.

Optionally, in the information processing method provided in the embodiment of the present application, determining, according to the N text information and the target classification model, category information corresponding to the N text information includes: word segmentation is carried out on the N text messages to obtain N segmented text messages; inputting the text information subjected to the N word segmentation into a target classification model for classification processing to obtain category information corresponding to the N text information.

For example, the object classification model described above may use the BERT model. In addition, some conversations are meaningless, such as "good, you say, i hear clearly," at this time, a simple judgment can be performed on the conversations through the scene classification module to judge whether the current sentence is meaningful, and at the same time, a simple classification is performed on the conversational content, such as classifying into credit cards, account blocking classes, loans, etc., so that subsequent information extraction is facilitated, and the classification of the conversational content can be set by oneself. And the scene classification module may include therein a BERT model (the object classification model described above). In classifying the dialogue content using the BERT model (the target classification model described above), text data (i.e., the text 1, the text 2, the text 3, etc.) after the data cleansing process may be segmented, and then the segmented text may be input to the BERT model (the target classification model described above) for classification processing, and the corresponding categories of the text data (i.e., the text 1, the text 2, the text 3, etc.) may be output.

Through the scheme, the dialogue voice content between the customer service and the customer can be classified rapidly and accurately.

Optionally, in the information processing method provided in the embodiment of the present application, the object classification model is obtained by: obtaining M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1; labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample; determining training data according to M segmented text samples and category information corresponding to each segmented text sample; and learning and training the original classification model by using training data to obtain a target classification model.

For example, in model training of the BERT model (the target classification model described above), training data may be prepared first, and the process of preparing the training data may be: for example, a sentence may be very short, such as "good, line", during a conversation, and very little information may be contained in the sentence. In order to make the model more accurate, the current sentence and the first two sentences can be spliced to be used as a text. If the current sentence is sentence3, the spliced sentence is sentence1+ sentence2+ sentence3. If the current dialogue is the first sentence or the second sentence, the current sentence is taken as the text, or the previous sentence is added as the text. After the splicing is completed, the text can be segmented by using the jieba model, and finally the segmented text is marked, such as credit cards, loans and business handling classes, and the classes can be set by themselves. And the final training data is in the format of (text, category labels).

For example, after the training data is prepared, a process of model training may be performed. For example, the BERT model (the target classification model) can be subjected to parameter fine adjustment, training data is input into the model, and the model can learn information in a text autonomously.

Through the scheme, the classification model can be conveniently subjected to learning training, so that the output result of the classification model is more accurate.

Optionally, in the information processing method provided in the embodiment of the present application, learning and training the original classification model by using training data, obtaining the target classification model includes: learning and training the original classification model by using training data to obtain a trained classification model; judging whether the trained classification model meets a preset stopping condition, wherein the stopping condition is a condition for stopping training the original classification model; if the trained classification model does not meet the stopping condition, continuing to learn and train the original classification model; and if the trained classification model meets the stopping condition, evaluating the trained classification model to obtain an evaluation result, and obtaining a target classification model based on the evaluation result.

For example, when model training is performed on the BERT model, it may be determined whether the BERT model after training reaches a stop condition (the above-described preset stop condition). And the stop condition (the above-described preset stop condition) may include: the number of training models reaches the designated training round n, the loss function value is smaller than the set value m, the accuracy rate change amplitude is smaller than the set value k%, and the training can be stopped when three conditions are met. When judging whether the trained BERT model reaches the stop condition (the preset stop condition), if the trained BERT model does not meet the three conditions (the number of training models reaches the appointed training round n, the loss function value is smaller than the set value m and the accuracy rate change amplitude is smaller than the set value k%), continuing to train the BERT model; if the trained BERT model meets any one of the three conditions (the number of times of training the model reaches the specified training round n, the loss function value is smaller than the set value m, and the accuracy rate change amplitude is smaller than the set value k%) in the example, the BERT model needs to be evaluated, and the finally trained BERT model is determined according to the model evaluation result.

By the scheme, whether the BERT model can be stopped to be trained can be quickly and accurately determined.

Optionally, in the information processing method provided in the embodiment of the present application, obtaining the target classification model based on the evaluation result includes: judging whether the evaluation result accords with the expected result; if the evaluation result does not accord with the expected result, adjusting the target data information to obtain adjusted target data information, wherein the target data information at least comprises: training data and parameter information of an original classification model; continuing to learn and train the original classification model based on the adjusted target data information; and if the evaluation result accords with the expected result, taking the trained classification model as a target classification model.

For example, in evaluating a model, the evaluation parameters may be accuracy, precision, recall. Furthermore, accuracy = number of samples for which all predictions are correct +.; the precision rate and recall rate are for each category, precision rate = number of correct predictions for a certain category ++model predicts as the total number of categories; recall = correct number of predictions for a category +.a total number of categories.

For example, when judging whether the evaluation result accords with the expectation, the current model can be saved as the final model according to three parameters of accuracy, precision and recall, if all the three parameters reach the expectation (the expectation result); if all three parameters are not expected (the expected result), that is, if any one of the three parameters is not expected (the expected result), the training data, the model parameters and the like are adjusted to continue training.

For example, the expected results may be 90% accuracy, 93% accuracy, and 95% recall. For example, when the model is evaluated, if the accuracy rate of the model obtained by evaluation is 92%, the accuracy rate reaches 95%, and the recall rate reaches 96%, the evaluation result of the model accords with the expected result; if the accuracy of the model obtained by evaluation is 85%, the precision reaches 95%, and the recall reaches 96%, since the accuracy of the model is 85%, and the accuracy of 85% does not reach 90% of the accuracy specified in the expected result, even if the other two parameters (precision and recall) reach the precision and recall specified in the expected result, the final result will be indicated as the evaluation result of the model does not conform to the expected result.

By the scheme, the model can be rapidly and accurately evaluated, so that a more accurate model can be trained.

Optionally, in the information processing method provided in the embodiment of the present application, inputting N pieces of text information and a target instruction into the information extraction model to process, and obtaining target information in the N pieces of text information includes: extracting information from the N pieces of text information according to the target instruction through an information extraction model to obtain information extracted from the N pieces of text information; and taking the information extracted from the N pieces of text information as target information in the N pieces of text information.

For example, key information (the target information described above) in a text (the N pieces of text information described above) may be extracted by the information extraction module. The input text of the model (the information extraction model described above) may be text data (i.e., the above-described text 1, text 2, text 3, etc.) subjected to the data cleansing process, and the key information (the above-described target information) may be extracted using the large model (the information extraction model described above). In addition, large models include, but are not limited to, GLM (generalized linear model, generalized Linear Models, for solving heterogeneous data analysis problems), GPT (generated Pre-Trained Transformer, a deep learning-based natural language processing model), and the like. And different instructions can be input into the model (the information extraction model described above) for different scenes. For example, for a credit card scenario, the instruction may be "extract name, card number, phone number" and for an account blocked scenario, the instruction may be "extract event keywords, no more than 10 words per keyword, no more than 5 keywords. The above is by way of example only, and the instructions may be self-setting. The input of the large model (information extraction model described above) is text and instructions, and the output is extraction information.

In summary, the key information in the dialogue content between the customer service and the customer can be extracted rapidly and accurately by using the large model and different instructions.

Optionally, in the information processing method provided in the embodiment of the present application, determining answer information corresponding to the question information based on the target information in the N pieces of text information and the information in the database includes: inputting target information in N pieces of text information into a search engine to obtain S pieces of information in a database, wherein the search engine is used for searching the information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than that between each piece of information in the T pieces of information and the target information, the T pieces of information are information except the S pieces of information in the database, S is a positive integer greater than 1, and T is a positive integer; and determining answer information corresponding to the question information based on the S pieces of information.

For example, the key information (the target information) extracted by the large model (the information extraction model) can be matched with the information in the database, for example, the client information can be acquired in the database system according to the name card number, and the current state of the card; based on event keywords such as account freeze, online banking malfunction, etc., a solution to the problem can be searched in the database. In addition, the search engine may use an elastic search (an open-source distributed search and analysis engine), and all the key information (the target information) extracted by using the large model (the information extraction model) is input into the elastic search (an open-source distributed search and analysis engine), and the engine returns TopK results with the highest similarity in the database. K can be set by oneself, and K can be set to 3, then these 3 results (above S pieces of information) that the similarity is highest in the database can be returned to customer service robot (above second object), then customer service robot (above second object) can select the most suitable result from these 3 results (above S pieces of information), and can regard it as final result (above answer information) the most suitable result that selects, and send final result (above answer information) to customer of financial institution (above first object), can realize real-time knowledge follow-up.

Through the scheme, the answer content corresponding to the client questions can be obtained rapidly and accurately according to the information stored in the database in advance.

For example, the large model is a new research result at present, and compared with the traditional model, the large model has strong generalization capability, has the capability of generating dialogue, can extract key information by context understanding dialogue content, and can improve the capability of information screening and matching to a greater extent.

For example, in order to improve the service efficiency of telephone customer service, the embodiment provides a knowledge following method based on an artificial intelligence technology, and through the combination of a traditional model and a large model, the method aims at understanding dialogue content, giving relevant service information in real time, and helping customer service to answer customer questions faster and better.

For example, fig. 2 is a schematic diagram of modules corresponding to an information processing method in an embodiment of the present application, and as shown in fig. 2, the modules corresponding to the information processing method in the embodiment of the present application include a 1 voice transcription module, a 2 data cleaning module, a 3 scene classification module, a 4 information extraction module, and a 5 search module. The content of each module is specifically as follows:

and 1, a voice transcription module. The embodiment is a real-time knowledge following method, so that the input voice is firstly transcribed in real time, and then the transcribed text is analyzed. Customer service and customer speech channels are different, and are used as distinction in transfer, customer service and customer speaking are alternately performed, and each character is transferred into a sentence. Namely, customer service: sense 1, client: sense 2, customer service: sendence 3.

2, a data cleaning module, which cleans the obtained text data, and mainly comprises three steps: (1) Stop word filtering, filtering words which have high occurrence times but have no significant meaning in text information, including imaginary words such as word and phrase, adverbs, preposition and the like, and high-frequency words such as' yes, no, and the like, wherein the stop words needing filtering are set by themselves. (2) removing numbers, english and special characters in the text. (3) replacing the special vocabulary. When the special vocabulary is transcribed, errors can occur in transcription, for example, an A1 bank can be transcribed into an A2 bank, and substitution processing is carried out on the important special vocabulary, so that the special vocabulary is set automatically.

And 3, a scene classification module. Because some conversations are meaningless, such as 'good, you say, i hear nothing clearly', the purpose of the module is to simply judge the conversations, judge whether the current sentence is meaningful or not, and simply classify the conversations, such as credit cards, account blocking classes, loans and the like, so that the subsequent information extraction is facilitated, and the classes can be set by themselves. The classification model uses the BERT model, and fig. 3 is a flowchart of training the classification model in the embodiment of the present application, and as shown in fig. 3, a specific flow of training the classification model is as follows:

Training data preparation 31. A sentence may be very short in the course of a conversation, such as "good, lines", and very little information is contained in the sentence. In order to make the model more accurate, the current sentence and the first two sentences are spliced to be used as texts. If the current sentence is sentence3, the spliced sentence is sentence1+ sentence2+ sentence3. If the current dialogue is the first sentence or the second sentence, the current sentence is taken as the text, or the previous sentence is added as the text. After the splicing is completed, the text is segmented by using a jieba model, and finally the segmented text is marked, such as credit cards, loans and business handling, and the category is set automatically. The final training data is in the format of (text, category labels).

Model 32 training. And performing parameter fine adjustment on the BERT model, inputting training data into the model, and enabling the model to autonomously learn information in the text.

33, judging: whether a stop condition is reached. The stop conditions include: the method achieves the purposes that the designated training round n is achieved, the loss function value is smaller than the set value m, and the accuracy rate change amplitude is smaller than the set value k%. And if any one of the three conditions is met, training is stopped.

34 model evaluation. The evaluation parameters are accuracy, precision and recall. Accuracy = number of samples that all predictions were correct +.total number of samples; the precision rate and recall rate are for each category, precision rate = number of correct predictions for a certain category ++model predicts as the total number of categories; recall = correct number of predictions for a category +.a total number of categories.

35, judging: whether or not it meets the expectations. According to the three parameters of the accuracy rate, the precision rate and the recall rate, if all the three parameters reach the expectations, the current model is stored as a final model; if any one of the three parameters does not meet the expectations, the training data, the model parameters and the like are adjusted, and training is continued.

And 4, an information extraction module. Information in the text is extracted. The input text is the text in the 3 scene classification module, and the large model is used for extracting information. The large model includes, but is not limited to, GLM (generalized linear model, generalized Linear Models for solving different kinds of data analysis problems), GPT (generated Pre-Trained Transformer, a deep learning-based natural language processing model), and the like, and different instructions are input for different scenes. For example, for a credit card scenario, the instruction is "extract name, card number, phone number", while for an account blocked scenario, the instruction is "extract event keywords, each keyword is no more than 10 words, no more than 5 keywords". The above is by way of example only, and the instructions are self-setting. The inputs to the large model are text and instructions, the outputs are extracted information, and the outputs are used in the subsequent modules.

And 5, searching the module. The module is linked with the database, and matches the information extracted by the 4 information extraction module with the information in the database, for example, the client information can be obtained in the system according to the name card number, and the current state of the card; based on event keywords such as account freeze, online banking malfunction, etc., a solution to the problem can be searched in the database. The search engine inputs all the information acquired by the 4 information extraction modules into the elastomer search (an open-source distributed search and analysis engine) by using the elastomer search (an open-source distributed search and analysis engine), and the engine returns TopK results with highest similarity in the database. And K can be set by oneself, usually set K to 3, return these 3 results to customer service, can realize the real-time knowledge and follow.

Therefore, the embodiment provides a more efficient and accurate knowledge follow-up scheme, which can help customer service to answer customer questions faster and better.

In summary, in the information processing method provided by the embodiment of the present application, N text information is obtained, where the N text information includes at least problem information related to a financial field, which is set by a first object to a second object, where the second object is an object in a financial institution that provides a financial service to the first object, and N is a positive integer greater than 1; determining category information corresponding to the N text information according to the N text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model; determining a target instruction according to category information corresponding to the N pieces of text information, inputting the N pieces of text information and the target instruction into an information extraction model for processing, and obtaining target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information; and determining answer information corresponding to the question information based on the target information in the N pieces of text information and the information in the database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information, so that the problem that in the related technology, the question related to finance is answered by a customer in a mode of manually inquiring related files through customer service, and the problem that the efficiency of answering the question is lower is solved. The method comprises the steps of obtaining a plurality of text messages at least comprising questions presented by clients to customer service, wherein the questions presented by the clients are questions related to finance, determining category information corresponding to the plurality of text messages according to the plurality of text messages and the classification model, determining instructions for extracting key information according to the category information corresponding to the plurality of text messages, inputting the plurality of text messages and the instructions into the information extraction model for processing, obtaining key information in the plurality of text messages, determining answers corresponding to the questions presented by the clients based on the key information in the plurality of text messages and information in a database, and returning the answers to the clients, so that the questions related to finance presented by the clients are answered in a mode of manually inquiring related files by the customer service is not needed, and further, the effect of improving the efficiency of answering the questions of the clients is achieved.

It should be noted that the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that illustrated herein.

The embodiment of the application also provides an information processing device, and it should be noted that the information processing device provided by the embodiment of the application can be used for executing the information processing method provided by the embodiment of the application. The information processing apparatus provided in the embodiment of the present application is described below.

Fig. 4 is a schematic diagram of an information processing apparatus provided according to an embodiment of the present application. As shown in fig. 4, the apparatus includes: a first acquisition unit 401, a first determination unit 402, a first processing unit 403, and a second processing unit 404.

Specifically, the first obtaining unit 401 is configured to obtain N pieces of text information, where the N pieces of text information at least include problem information related to a financial field, which is set forth by a first object to a second object, where the second object is an object in a financial institution that provides a financial service to the first object, and N is a positive integer greater than 1;

A first determining unit 402, configured to determine category information corresponding to the N text information according to the N text information and a target classification model, where the target classification model is a model constructed based on a deep learning model;

the first processing unit 403 is configured to determine a target instruction according to category information corresponding to the N pieces of text information, and input the N pieces of text information and the target instruction into the information extraction model for processing, so as to obtain target information in the N pieces of text information, where the target instruction is an instruction for extracting information from the N pieces of text information;

the second processing unit 404 is configured to determine answer information corresponding to the question information based on the target information in the N text information and information in a database, and return the answer information to the first object, where the database is configured to store information related to the question information.

In summary, in the information processing apparatus provided in the embodiments of the present application, N pieces of text information are acquired by the first acquiring unit 401, where the N pieces of text information at least include problem information related to a financial field, which is set by a first object to a second object, where the second object is an object in a financial institution that provides a financial service to the first object, and N is a positive integer greater than 1; the first determining unit 402 determines category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, where the target classification model is a model constructed based on a deep learning model; the first processing unit 403 determines a target instruction according to category information corresponding to the N pieces of text information, and inputs the N pieces of text information and the target instruction into the information extraction model to process, so as to obtain target information in the N pieces of text information, where the target instruction is an instruction for extracting information from the N pieces of text information; the second processing unit 404 determines answer information corresponding to the question information based on the target information in the N text information and information in the database, and returns the answer information to the first object, where the database is used to store information related to the question information, which solves the problem in the related art that the problem related to finance is answered by a customer by manually querying the related file through customer service, resulting in a problem that the efficiency of answering the problem is lower. The method comprises the steps of obtaining a plurality of text messages at least comprising questions presented by clients to customer service, wherein the questions presented by the clients are questions related to finance, determining category information corresponding to the plurality of text messages according to the plurality of text messages and the classification model, determining instructions for extracting key information according to the category information corresponding to the plurality of text messages, inputting the plurality of text messages and the instructions into the information extraction model for processing, obtaining key information in the plurality of text messages, determining answers corresponding to the questions presented by the clients based on the key information in the plurality of text messages and information in a database, and returning the answers to the clients, so that the questions related to finance presented by the clients are answered in a mode of manually inquiring related files by the customer service is not needed, and further, the effect of improving the efficiency of answering the questions of the clients is achieved.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the second processing unit includes: the first input module is used for inputting target information in N pieces of text information into the search engine to obtain S pieces of information in the database, wherein the search engine is used for searching information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than the similarity between each piece of information in the T pieces of information and the target information, the T pieces of information are the information except the S pieces of information in the database, S is a positive integer greater than 1, and T is a positive integer; and the first determining module is used for determining answer information corresponding to the question information based on the S pieces of information.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the first processing unit includes: the first extraction module is used for extracting information from the N pieces of text information according to the target instruction through the information extraction model to obtain information extracted from the N pieces of text information; and the second determining module is used for taking information extracted from the N pieces of text information as target information in the N pieces of text information.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the first determining unit includes: the first processing module is used for carrying out word segmentation processing on the N text messages to obtain N segmented text messages; the second input module is used for inputting the text information subjected to the N word segmentation into the target classification model for classification processing to obtain category information corresponding to the N text information.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the object classification model is obtained by: the second acquisition unit is used for acquiring M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1; the first labeling unit is used for labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample; the second determining unit is used for determining training data according to the M segmented text samples and category information corresponding to each segmented text sample; the first training unit is used for learning and training the original classification model by utilizing training data to obtain a target classification model.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the first training unit includes: the first training module is used for learning and training the original classification model by utilizing training data to obtain a trained classification model; the first judging module is used for judging whether the trained classification model meets preset stopping conditions, wherein the stopping conditions are conditions for stopping training the original classification model; the second training module is used for continuing to learn and train the original classification model if the trained classification model does not meet the stop condition; and the second processing module is used for evaluating the trained classification model to obtain an evaluation result if the trained classification model meets the stop condition, and obtaining a target classification model based on the evaluation result.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the second processing module includes: the first judging sub-module is used for judging whether the evaluation result accords with the expected result or not; the first adjustment sub-module is configured to adjust the target data information if the evaluation result does not conform to the expected result, so as to obtain adjusted target data information, where the target data information at least includes: training data and parameter information of an original classification model; the first training sub-module is used for continuously learning and training the original classification model based on the adjusted target data information; and the first determination submodule is used for taking the trained classification model as a target classification model if the evaluation result accords with the expected result.

Optionally, in the information processing apparatus provided in the embodiment of the present application, the first acquisition unit includes: the first acquisition module is used for acquiring dialogue voice information of a dialogue between the first object and the second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object; the third processing module is used for carrying out transcription processing on the dialogue voice information to obtain N original text information; the fourth processing module is used for preprocessing the N original text messages to obtain N text messages, wherein the preprocessing is at least one of the following: filtering stop words, removing target characters and replacing target words.

The information processing apparatus includes a processor and a memory, and the first acquisition unit 401, the first determination unit 402, the first processing unit 403, the second processing unit 404, and the like described above are stored in the memory as program units, and the processor executes the program units stored in the memory to realize the corresponding functions.

The processor includes a kernel, and the kernel fetches the corresponding program unit from the memory. The kernel can be provided with one or more than one, and the efficiency of answering the client questions is improved by adjusting the kernel parameters.

The memory may include volatile memory, random Access Memory (RAM), and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM), among other forms in computer readable media, the memory including at least one memory chip.

An embodiment of the present invention provides a computer-readable storage medium having stored thereon a program which, when executed by a processor, implements the information processing method.

The embodiment of the invention provides a processor for running a program, wherein the information processing method is executed when the program runs.

As shown in fig. 5, an embodiment of the present invention provides an electronic device, where the device includes a processor, a memory, and a program stored in the memory and executable on the processor, and when the processor executes the program, the following steps are implemented: acquiring N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object which provides financial services for the first object in a financial institution, and N is a positive integer greater than 1; determining category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model; determining a target instruction according to category information corresponding to the N pieces of text information, and inputting the N pieces of text information and the target instruction into an information extraction model for processing to obtain target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information; and determining answer information corresponding to the question information based on target information in the N pieces of text information and information in a database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information.

The processor also realizes the following steps when executing the program: determining answer information corresponding to the question information based on the target information in the N text information and the information in the database comprises: inputting target information in the N pieces of text information into a search engine to obtain S pieces of information in the database, wherein the search engine is used for searching information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than the similarity between each piece of information in the T pieces of information and the target information, the T pieces of information are information except the S pieces of information in the database, S is a positive integer greater than 1, and T is a positive integer; and determining answer information corresponding to the question information based on the S pieces of information.

The processor also realizes the following steps when executing the program: inputting the N text messages and the target instructions into an information extraction model for processing, and obtaining target information in the N text messages comprises the following steps: extracting information from the N pieces of text information according to the target instruction through the information extraction model to obtain information extracted from the N pieces of text information; and taking the information extracted from the N pieces of text information as the target information in the N pieces of text information.

The processor also realizes the following steps when executing the program: determining category information corresponding to the N text information according to the N text information and the target classification model comprises the following steps: performing word segmentation processing on the N text messages to obtain N segmented text messages; inputting the text information subjected to the N word segmentation into the target classification model for classification processing to obtain category information corresponding to the N text information.

The processor also realizes the following steps when executing the program: the target classification model is obtained by the following steps: obtaining M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1; labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample; determining training data according to the M segmented text samples and category information corresponding to each segmented text sample; and learning and training the original classification model by utilizing the training data to obtain the target classification model.

The processor also realizes the following steps when executing the program: learning and training the original classification model by using the training data, and obtaining the target classification model comprises the following steps: learning and training the original classification model by utilizing the training data to obtain a trained classification model; judging whether the trained classification model meets a preset stopping condition, wherein the stopping condition is a condition for stopping training the original classification model; if the trained classification model does not meet the stopping condition, continuing to learn and train the original classification model; and if the trained classification model meets the stopping condition, evaluating the trained classification model to obtain an evaluation result, and obtaining the target classification model based on the evaluation result.

The processor also realizes the following steps when executing the program: obtaining the target classification model based on the evaluation result includes: judging whether the evaluation result accords with an expected result or not; if the evaluation result does not accord with the expected result, adjusting the target data information to obtain adjusted target data information, wherein the target data information at least comprises: the training data and the parameter information of the original classification model; continuing to learn and train the original classification model based on the adjusted target data information; and if the evaluation result accords with the expected result, taking the trained classification model as the target classification model.

The processor also realizes the following steps when executing the program: the obtaining of the N text messages comprises the following steps: acquiring dialogue voice information of dialogue between the first object and the second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object; performing transcription processing on the dialogue voice information to obtain N original text information; preprocessing the N original text messages to obtain the N text messages, wherein the preprocessing is at least one of the following steps: filtering stop words, removing target characters and replacing target words.

The device herein may be a server, PC, PAD, cell phone, etc.

The present application also provides a computer program product adapted to perform, when executed on a data processing device, a program initialized with the method steps of: acquiring N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object which provides financial services for the first object in a financial institution, and N is a positive integer greater than 1; determining category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model; determining a target instruction according to category information corresponding to the N pieces of text information, and inputting the N pieces of text information and the target instruction into an information extraction model for processing to obtain target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information; and determining answer information corresponding to the question information based on target information in the N pieces of text information and information in a database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: determining answer information corresponding to the question information based on the target information in the N text information and the information in the database comprises: inputting target information in the N pieces of text information into a search engine to obtain S pieces of information in the database, wherein the search engine is used for searching information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than the similarity between each piece of information in the T pieces of information and the target information, the T pieces of information are information except the S pieces of information in the database, S is a positive integer greater than 1, and T is a positive integer; and determining answer information corresponding to the question information based on the S pieces of information.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: inputting the N text messages and the target instructions into an information extraction model for processing, and obtaining target information in the N text messages comprises the following steps: extracting information from the N pieces of text information according to the target instruction through the information extraction model to obtain information extracted from the N pieces of text information; and taking the information extracted from the N pieces of text information as the target information in the N pieces of text information.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: determining category information corresponding to the N text information according to the N text information and the target classification model comprises the following steps: performing word segmentation processing on the N text messages to obtain N segmented text messages; inputting the text information subjected to the N word segmentation into the target classification model for classification processing to obtain category information corresponding to the N text information.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: the target classification model is obtained by the following steps: obtaining M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1; labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample; determining training data according to the M segmented text samples and category information corresponding to each segmented text sample; and learning and training the original classification model by utilizing the training data to obtain the target classification model.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: learning and training the original classification model by using the training data, and obtaining the target classification model comprises the following steps: learning and training the original classification model by utilizing the training data to obtain a trained classification model; judging whether the trained classification model meets a preset stopping condition, wherein the stopping condition is a condition for stopping training the original classification model; if the trained classification model does not meet the stopping condition, continuing to learn and train the original classification model; and if the trained classification model meets the stopping condition, evaluating the trained classification model to obtain an evaluation result, and obtaining the target classification model based on the evaluation result.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: obtaining the target classification model based on the evaluation result includes: judging whether the evaluation result accords with an expected result or not; if the evaluation result does not accord with the expected result, adjusting the target data information to obtain adjusted target data information, wherein the target data information at least comprises: the training data and the parameter information of the original classification model; continuing to learn and train the original classification model based on the adjusted target data information; and if the evaluation result accords with the expected result, taking the trained classification model as the target classification model.

When executed on a data processing device, is further adapted to carry out a program initialized with the method steps of: the obtaining of the N text messages comprises the following steps: acquiring dialogue voice information of dialogue between the first object and the second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object; performing transcription processing on the dialogue voice information to obtain N original text information; preprocessing the N original text messages to obtain the N text messages, wherein the preprocessing is at least one of the following steps: filtering stop words, removing target characters and replacing target words.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, etc., such as Read Only Memory (ROM) or flash RAM. Memory is an example of a computer-readable medium.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises an element.

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and changes may be made to the present application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. which are within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.

Claims

1. An information processing method, characterized by comprising:

acquiring N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object which provides financial services for the first object in a financial institution, and N is a positive integer greater than 1;

determining category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model;

determining a target instruction according to category information corresponding to the N pieces of text information, and inputting the N pieces of text information and the target instruction into an information extraction model for processing to obtain target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information;

and determining answer information corresponding to the question information based on target information in the N pieces of text information and information in a database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information.

2. The method of claim 1, wherein determining answer information corresponding to the question information based on target information in the N text information and information in a database comprises:

inputting target information in the N pieces of text information into a search engine to obtain S pieces of information in the database, wherein the search engine is used for searching information from the database, the similarity between each piece of information in the S pieces of information and the target information is higher than the similarity between each piece of information in the T pieces of information and the target information, the T pieces of information are information except the S pieces of information in the database, S is a positive integer greater than 1,

t is a positive integer;

and determining answer information corresponding to the question information based on the S pieces of information.

3. The method of claim 1, wherein processing the N text messages and the target instruction input information extraction model to obtain target information in the N text messages comprises:

extracting information from the N pieces of text information according to the target instruction through the information extraction model to obtain information extracted from the N pieces of text information;

And taking the information extracted from the N pieces of text information as the target information in the N pieces of text information.

4. The method of claim 1, wherein determining category information corresponding to the N text information from the N text information and a target classification model comprises:

performing word segmentation processing on the N text messages to obtain N segmented text messages;

inputting the text information subjected to the N word segmentation into the target classification model for classification processing to obtain category information corresponding to the N text information.

5. The method of claim 1, wherein the object classification model is obtained by:

obtaining M text samples, and performing word segmentation on the M text samples to obtain M segmented text samples, wherein M is a positive integer greater than 1;

labeling the category of each segmented text sample to obtain category information corresponding to each segmented text sample;

determining training data according to the M segmented text samples and category information corresponding to each segmented text sample;

and learning and training the original classification model by utilizing the training data to obtain the target classification model.

6. The method of claim 5, wherein learning training the original classification model with the training data to obtain the target classification model comprises:

learning and training the original classification model by utilizing the training data to obtain a trained classification model;

judging whether the trained classification model meets a preset stopping condition, wherein the stopping condition is a condition for stopping training the original classification model;

if the trained classification model does not meet the stopping condition, continuing to learn and train the original classification model;

and if the trained classification model meets the stopping condition, evaluating the trained classification model to obtain an evaluation result, and obtaining the target classification model based on the evaluation result.

7. The method of claim 6, wherein deriving the object classification model based on the evaluation result comprises:

judging whether the evaluation result accords with an expected result or not;

if the evaluation result does not accord with the expected result, adjusting the target data information to obtain adjusted target data information, wherein the target data information at least comprises: the training data and the parameter information of the original classification model;

Continuing to learn and train the original classification model based on the adjusted target data information;

and if the evaluation result accords with the expected result, taking the trained classification model as the target classification model.

8. The method of claim 1, wherein obtaining N pieces of text information comprises:

acquiring dialogue voice information of dialogue between the first object and the second object, wherein the dialogue voice information at least comprises problem information related to the financial field, which is proposed by the first object to the second object;

performing transcription processing on the dialogue voice information to obtain N original text information;

preprocessing the N original text messages to obtain the N text messages, wherein the preprocessing is at least one of the following steps: filtering stop words, removing target characters and replacing target words.

9. An information processing apparatus, characterized by comprising:

the first obtaining unit is used for obtaining N pieces of text information, wherein the N pieces of text information at least comprise problem information related to the financial field, which is proposed by a first object to a second object, the second object is an object which provides financial services for the first object in a financial institution, and N is a positive integer greater than 1;

The first determining unit is used for determining category information corresponding to the N pieces of text information according to the N pieces of text information and a target classification model, wherein the target classification model is a model constructed based on a deep learning model;

the first processing unit is used for determining a target instruction according to category information corresponding to the N pieces of text information, inputting the N pieces of text information and the target instruction into an information extraction model for processing, and obtaining target information in the N pieces of text information, wherein the target instruction is an instruction for extracting information from the N pieces of text information;

and the second processing unit is used for determining answer information corresponding to the question information based on the target information in the N pieces of text information and information in a database, and returning the answer information to the first object, wherein the database is used for storing information related to the question information.

10. A computer-readable storage medium storing a program, wherein the program executes the information processing method according to any one of claims 1 to 8.

11. An electronic device comprising one or more processors and a memory for storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the information processing method of any of claims 1-8.