CN112989001B - Question and answer processing method and device, medium and electronic equipment - Google Patents
Question and answer processing method and device, medium and electronic equipment Download PDFInfo
- Publication number
- CN112989001B CN112989001B CN202110349133.XA CN202110349133A CN112989001B CN 112989001 B CN112989001 B CN 112989001B CN 202110349133 A CN202110349133 A CN 202110349133A CN 112989001 B CN112989001 B CN 112989001B
- Authority
- CN
- China
- Prior art keywords
- questions
- question
- user
- stored
- candidate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 26
- 238000000034 method Methods 0.000 claims abstract description 46
- 238000012549 training Methods 0.000 claims description 49
- 238000003062 neural network model Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 7
- 230000008685 targeting Effects 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 15
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 239000002609 medium Substances 0.000 description 13
- 230000003993 interaction Effects 0.000 description 6
- 238000010276 construction Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 238000002372 labelling Methods 0.000 description 4
- 238000012216 screening Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000002457 bidirectional effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000003058 natural language processing Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 239000012120 mounting media Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000008707 rearrangement Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/338—Presentation of query results
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The embodiment of the application discloses a question and answer processing method, a question and answer processing device, a medium and electronic equipment. Relates to the field of artificial intelligence, the method comprises the following steps: performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions; performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem; selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions; wherein the first feature is different from the second feature. By executing the scheme, the user questions can be accurately identified, the accuracy of user question answering is improved, and further user experience is improved.
Description
Technical Field
The embodiment of the application relates to the field of artificial intelligence, in particular to a question-answering processing method, a question-answering processing device, a question-answering processing medium and electronic equipment.
Background
With the continuous development of society, more and more people acquire knowledge and information through a question and answer system. The question-answering system (Question Answering System, QA) is a high-level form of information retrieval system that can answer questions posed by a user in natural language in accurate, compact natural language. The question-answering system is a research direction which is attracting attention and has wide development prospect in the fields of artificial intelligence and natural language processing.
Currently, industry typically employs a question and answer system built based on FAQ (Frequent Asked Questions) question and answer technology to process questions of users, such as a search-based method TF-IDF (Term Frequency-Inverse Document Frequency), a statistical machine learning-based method, and a deep learning-based method DSSM (Deep Structured Semantic Model) to return answers to questions of users. However, the above method only considers the case that only a single question is included in the question of the user, and when a plurality of associated questions are included in the question of the user, it is difficult to completely match all the questions and give an accurate answer. In an actual user interaction scenario, the data rate of more than 2 questions included in the user question is 39%, so it is important to be able to identify the number of questions included in the user question and return a corresponding answer to each question.
Disclosure of Invention
The embodiment of the application provides a question and answer processing method, a question and answer processing device, a medium and electronic equipment, which can be used for identifying the number of questions included in a user question and returning corresponding answers to each question, so that the purposes of improving the answer accuracy of a question and answer system and the intelligence of the question and answer system are achieved.
In a first aspect, an embodiment of the present application provides a question-answering processing method, where the method includes:
Performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions;
performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem;
selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions; wherein the first feature is different from the second feature.
In a second aspect, an embodiment of the present application provides a question-answering processing apparatus, where the apparatus includes:
the initial question determining module is used for performing first characteristic matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions;
the candidate problem determining module is used for performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem;
a target question determining module, configured to select a target question from the candidate questions according to third feature data between the user question and the candidate questions, and take an answer associated with the target question as an answer of the user question; wherein the first feature is different from the second feature.
In a third aspect, embodiments of the present application provide a computer-readable storage medium having stored thereon a computer program that, when executed by a processor, implements a question-answering method according to embodiments of the present application.
In a fourth aspect, an embodiment of the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and capable of being executed by the processor, where the processor executes the computer program to implement a question-answering processing method according to an embodiment of the present application.
According to the technical scheme provided by the embodiment of the application, the user questions are matched with the questions in the pre-stored question-answering pair, the answers of the user questions are determined in the known question answers, the user questions are subjected to characteristic matching of different layers and different types for a plurality of times with the pre-stored questions in the pre-stored question-answering pair, the searching range of the answers of the user questions is gradually reduced, the pre-stored questions with the highest matching degree with the user questions are determined as target questions, the answers associated with the target questions are used as the answers of the user questions, the answers of each sub-question in the user questions are ensured to be corresponding through the characteristic matching of multiple layers and multiple angles, the answer accuracy of a question-answering system and the intelligence of the question-answering system are improved, and further the user experience is improved.
Drawings
Fig. 1 is a flowchart of a question-answering processing method according to an embodiment of the present application;
FIG. 2 is a flowchart of another question-answering processing method according to the second embodiment of the present application;
FIG. 3 is a flowchart of yet another question-answering method provided in the third embodiment of the present application;
FIG. 4 is a flowchart of yet another question-answering method according to the fourth embodiment of the present application;
fig. 5 is a schematic structural diagram of a question-answering processing device provided in a fifth embodiment of the present application;
fig. 6 is a schematic structural diagram of an electronic device according to a seventh embodiment of the present application.
Detailed Description
The present application is described in further detail below with reference to the drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the application and not limiting thereof. It should be further noted that, for convenience of description, only some, but not all of the structures related to the present application are shown in the drawings.
Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although a flowchart depicts steps as a sequential process, many of the steps may be implemented in parallel, concurrently, or with other steps. Furthermore, the order of the steps may be rearranged. The process may be terminated when its operations are completed, but may have additional steps not included in the figures. The processes may correspond to methods, functions, procedures, subroutines, and the like.
Example 1
Fig. 1 is a flowchart of a question-answering processing method provided in an embodiment of the present application, where the embodiment may be applicable to a case where a human-computer interaction question-answering system receives a user question and feeds back a matched answer to the user aiming at the user question. The method can be executed by the question and answer processing device provided by the embodiment of the application, and the device can be realized by software and/or hardware and can be integrated into an electronic device running the system.
As shown in fig. 1, the question-answering processing method includes:
s110, performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions.
The user questions refer to questions to be answered received by the human-computer interaction question answering system. The pre-stored question-answer pair is composed of a pre-stored question and an answer for answering the pre-stored question. The prestored question-answer pairs are sets of common questions and answers which are arranged by related technicians and are stored in a local database or a cloud end of the man-machine interaction question-answer system in advance. In the case that the user questions are known and the pre-stored question-answer pair exists, the answers of the pre-stored questions successfully matched can be used as the answers of the user questions by matching the user questions with the pre-stored questions in the pre-stored question-answer pair.
And carrying out first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions. The initial question is an initial question successfully matched with the first feature of the user question, and the preliminary screening of the pre-stored questions is realized by carrying out the first feature matching, so that the searching range of answers of the user questions is narrowed.
In an alternative embodiment, the first characteristic refers to the frequency of occurrence in the pre-stored questions of the linguistic units constituting the user questions, the linguistic units comprising words or terms.
And carrying out first feature matching on the user questions and the pre-stored questions, specifically, counting the frequency of single words or words of language units forming the user questions in the pre-stored questions, wherein the higher the frequency of the same words or words in the user questions and the pre-stored questions is, the higher the possibility that the user questions and the pre-stored questions correspond to the same answer is. In order to reduce the influence of word splitting errors on the determination of the initial problem in word segmentation processing of the user problem, preferably, the words are used as language units, that is, the frequency of occurrence of single words of the user problem in the pre-stored problem is counted. By so doing, the search range of answers for answering user questions can be quickly determined, and the processing efficiency of user questions is improved.
S120, performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem.
The candidate questions refer to the result of further screening of the initial questions, and the candidate questions are pre-stored questions matched through the first characteristic and the second characteristic. The first feature match literally determines an approximate search range for the user's question answer, and the second feature is different from the first feature, the second feature match being a deeper feature match relative to the first feature match. Since sentences obtained by different combinations of the same text have differences in meaning, analysis of deep features of user problems is also required. Optionally, the second feature is a semantic feature. And matching semantic features of the user questions and the initial questions, and taking the initial questions successfully matched as candidate questions, so that the pre-stored questions are screened again, and the searching range of the answers of the user questions is further narrowed. The candidate questions are more similar to the user questions than the initial questions, not only in language units, but also closer to the user questions in sentence meaning.
S130, selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions.
After the first feature match and the second feature match, a final search range of the user question answers, i.e., the answers to the candidate questions, is determined. As a user problem may include sub-problems, for example, in a banking scenario, the user problem is "how the interest rate of the product is high, and the risk is high", and the user problem includes two sub-problems of the interest rate of the financial product and the risk of the financial product. Further analysis of sentence features such as sentence structure and sentence meaning of the user question itself is also required in order to ensure that the user question gets a complete answer.
And selecting a target question from the candidate questions according to third characteristic data between the user question and the candidate questions. Specifically, the sentence characteristics of the user questions and the candidate questions are combined with the matching degree between the sentence meanings of the user questions and the candidate questions, the target questions are finally determined, and the answers associated with the target questions are used as answers of the user questions. The target questions can be regarded as different expression modes of the user questions, and the target questions and the user questions correspond to the same answer.
In an alternative embodiment, the third characteristic data includes: syntactic features, contextual features, and semantic similarity features.
In an alternative embodiment, the syntactic characteristics may be determined by: analyzing the user problem and the candidate problem by utilizing dependency syntax respectively to obtain syntax structure information; wherein the syntax structure information is information associated with semantics; extracting sentence component information and sentence component combination relations of the user questions and the candidate questions respectively; and vectorizing the syntax structure information according to the number of the sentence components and the sentence component combination relations, and splicing vectorized results to serve as syntax features.
The dependency syntax is to analyze sentences into a dependency syntax tree, and describe the dependency relation among various words. That is, a syntactically collocation relationship between words is indicated, which is semantically associated. Illustratively, the sentence "conference announces a first list of senior yards. The words "announce" govern "meeting", "have" and "list" are available through dependency syntax, so these governments can be used as collocations for "announce". And respectively analyzing the user problem and the candidate problem by utilizing the dependency syntax to obtain the syntax structure information.
Extracting sentence component information and sentence component combination relation of a user problem and a candidate problem respectively, wherein the sentence component information refers to composition components of sentences, and the sentence components comprise at least one of the following: subject, predicate, object, animal, subject, idiom, complement, and center. The sentence component combination relationship includes at least one of: a master-slave relationship or a moving guest relationship. And vectorizing the syntax structure information according to the number of sentence components and the number of sentence component combination relations, and splicing vectorized results to serve as syntax features. The sentence component and sentence component combination relationship may reflect the number of sub-questions included in the user question.
In an alternative embodiment, the contextual characteristics may be determined by: extracting context semantic information of the user problem and the candidate problem respectively; and vectorizing the context semantic information respectively, and splicing vectorization results to serve as the semantic features.
Where the contextual features may reflect sentence semantic features as a whole. Since the words constituting the sentence do not exist independently, there is an association relationship between the words. The semantic features of the sentence are grasped as a whole, and then the association relationship between the words needs to be considered. Optionally, the text contents of the user problem and the candidate problem are encoded by using a bidirectional LSTM (Long-Short Term Memory, long-short term memory model), the context semantic information of the user problem and the candidate problem is extracted respectively, then the context semantic information is vectorized respectively, and the vectorized results are spliced to be used as semantic features.
The semantic similarity feature refers to an index for measuring the similarity between the user problem and the candidate problem at the semantic level, and in an optional embodiment, the semantic similarity feature is the sentence-to-semantic similarity score output by the semantic similarity model.
According to the technical scheme provided by the embodiment of the application, the user questions are matched with the questions in the pre-stored question-answering pair, the answers of the user questions are determined in the known question answers, the user questions are subjected to characteristic matching of different layers and different types for a plurality of times with the pre-stored questions in the pre-stored question-answering pair, the searching range of the answers of the user questions is gradually reduced, the pre-stored questions with the highest matching degree with the user questions are determined as target questions, the answers associated with the target questions are used as the answers of the user questions, the answers of each sub-question in the user questions are ensured to be corresponding through the characteristic matching of multiple layers and multiple angles, the answer accuracy of a question-answering system and the intelligence of the question-answering system are improved, and further the user experience is improved.
Example two
Fig. 2 is a flowchart of another question-answering processing method according to the second embodiment of the present application. The present embodiment is further optimized on the basis of the above embodiment. Specifically, the method includes performing first feature matching on a user question and a pre-stored question in a pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions, including: dividing the user problem into independent language units; calculating a first similarity score of the user problem and the pre-stored problem according to the occurrence frequency of each pre-stored problem in the pre-stored question-answering pair of each language unit; and selecting at least two pre-stored questions with the first similarity score larger than a preset similarity threshold as initial questions.
As shown in fig. 2, the question-answering processing method includes:
s210, dividing the user problem into independent language units.
The language unit is a basic unit for forming a user problem, and optionally, the language unit is a word or a single word. To avoid word splitting errors, it is preferable to use a single word as the linguistic unit. That is, the user question is split into individual words.
S220, calculating a first similarity score of the user problem and the pre-stored problem according to the occurrence frequency of each language unit in each pre-stored problem in the pre-stored question-answering pair.
Optionally, the frequency of occurrence of each word in the user question in each pre-stored question in the pre-stored question-answer pair is directly used as the first similarity score. Preferably, the frequency of each character in the user questions in each pre-stored question and answer pair and the number of the pre-stored questions in the pre-stored question and answer pair are counted, and information such as the average length of the pre-stored questions in the pre-stored question and answer pair is calculated comprehensively, so that the first similarity score of the user questions and each pre-stored question is calculated comprehensively.
S230, selecting at least two pre-stored questions with the first similarity score larger than a preset similarity threshold as initial questions.
The preset similarity threshold is an empirical value determined by a related technician according to actual situations, which is not limited herein, and is determined specifically according to actual situations. The higher the first similarity score, the higher the degree of matching of the user question and the pre-stored question. And taking at least two pre-stored questions with the first similarity score being larger than a preset similarity threshold as initial questions.
In an alternative embodiment, calculating a first similarity score between the user question and each pre-stored question in the pre-stored question-answering pair according to the occurrence frequency of each language unit, including:
calculating a first similarity score of the user question and the pre-stored question according to the following formula:
wherein t is i Representing language units in the user question faq j Representing the pre-stored problem, f (t i ,faq j ) Representing t i Faq on pre-stored problem j Frequency of occurrence k 1 And b is a first adjustment coefficient and a second adjustment coefficient, w i Represents the correlation weight, avgFAQ represents the average word length of the pre-stored questions, N represents the number of the pre-stored question-answer pairs, N (t) i ) The representation includes t i Faq of the pre-stored problem j Number, p i Is used as a weight to represent t i Importance level. Wherein k is 1 And b is an experience value determined by a relevant technician according to actual conditions, n is the number of the language units in the user problem, and i is the number of the language units in the user problem. Subscript j represents the jth pre-stored question faq in the pre-stored question-answer pair j 。
Specifically, it can be determined according to t i Species determination p of (2) i For example, the real word weight is determined to be 2, the stop word weight is determined to be 0.1, and the other types of weights are determined to be 0.1, so that the real word is highlighted and the stop word is weakened.
S240, performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem.
S250, selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions.
According to the technical scheme provided by the embodiment of the application, the user problem is divided into independent language units. According to the occurrence frequency of each pre-stored problem in the pre-stored question-answering pair of each language unit, calculating a first similarity score of the user problem and the pre-stored problems, and selecting at least two pre-stored problems with the first similarity score larger than a preset similarity threshold as initial problems. By executing the scheme, the search range of the answers for answering the user questions can be rapidly determined, the preliminary screening of the pre-stored questions is realized, the search range of the answers of the user questions is reduced, and the processing efficiency of the user questions is improved.
Example III
Fig. 3 is a flowchart of yet another question-answering processing method according to the third embodiment of the present application. The present embodiment is further optimized on the basis of the above embodiment. Specifically, the method includes performing a second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem, including: judging whether the initial questions and the user questions form semantic similar sentence pairs or not by utilizing a semantic similarity model according to the user questions, the text content information and sentence structure information of each initial question; if so, the initial question is determined to be a candidate question.
As shown in fig. 3, the question-answering processing method includes:
and S310, performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions.
S320, judging whether the initial questions and the user questions can form semantic similar sentence pairs by utilizing a semantic similarity model according to the user questions, the text content information and sentence structure information of each initial question.
S330, if yes, determining the initial problem as a candidate problem.
The semantic similar sentence pair refers to a sentence pair formed by two semantically similar sentences, the two sentences forming the sentence pair are expressed in different modes, and the user questions and the initial questions forming the semantic similar sentence pair can correspond to the same answer.
The semantic similarity model is used for calculating semantic similarity between the user problem and each initial problem, and determining whether the input user problem and initial problem can form a semantic similar sentence pair according to the semantic similarity. The semantic similarity model is a pre-training completion model, text content information and sentence structure information of the user problem and each initial problem are input into the semantic similarity model, and the semantic similarity model outputs semantic similarity scores of the user problem and the initial problem and a judgment result of whether the user problem and the initial problem form a semantic similar sentence pair or not.
The semantic similarity model takes semantic similarity judgment as a classification problem, determines the input formed by the user problem and the initial problem, the probability of belonging to the semantic similarity sentence pair and the probability of not belonging to the semantic similarity sentence pair, and compares the two probabilities to obtain a judgment result of whether the semantic similarity sentence pair is similar. In so doing, the problem of similarity calculation is effectively solved from the perspective of deep semantics.
In an optional embodiment, before determining whether the initial question and the user question can form a semantic similar sentence pair according to the text content information and sentence structure information of the user question and each initial question by using a semantic similarity model, the method further includes a training process of the semantic similarity model: determining the label data of training sample sentence pairs by utilizing a pre-trained semantic similar sentence pair judging model; wherein the tag data includes: semantic similarity sentences score classification attributes and sentence-to-semantic similarity;
The training sample sentence pairs are formed by splicing two problems, and the semantic similar sentence pair judgment model is a model for determining label data of the training sample sentence pairs. The tag data includes: semantic similar sentences score classification attributes and sentence-to-semantic similarity. Extracting the text content information and the sentence structure information of the training sample sentence pair as feature data; and training the semantic similarity model by using the feature data and the tag data as training data so as to enable the semantic similarity model to output semantic similarity scores of semantic similarity sentences and classification attributes and sentence-to-semantic similarity scores.
The semantic similar sentence pair belongs to a training sample sentence pair, and the training sample sentence pair does not belong to two types of semantic similar sentence pairs. The semantic similarity score is the probability that the training sample sentence pair belongs to the semantic similarity sentence pair, or the probability that the training sample sentence pair does not belong to the semantic similarity sentence pair, and the probability is alternatively used as the semantic similarity score.
And training the semantic similarity model by taking the text content information and the sentence structure information of the extracted training sample sentence pairs as characteristic data and taking the characteristic data and the tag data as training data of the semantic similarity model. Optionally, the semantic similarity model is a BERT (Bidirectional Encoder Representation from Transformers) semantic similarity model.
To enable automated determination of training sample tags, in an alternative embodiment, a semantic similarity sentence pair judgment model is constructed to enable automated determination of tag data of training samples required to develop a semantic similarity model.
The construction process of the training sample of the semantic similar sentence pair judgment model is as follows: performing the first feature matching on the two sample problems in the training sample sentence pair, and taking the training sample sentence pair as a positive sample sentence pair if the matching is successful; and if the matching fails, taking the training sample sentence pair as a negative example sample sentence. The output of the semantic similarity sentence pair judgment model is label data of a semantic similarity model training sample, namely semantic similarity sentences score classification attribute and sentence pair semantic similarity. Optionally, the semantic similar sentence pair judgment model is an LTR (Learning to Rank) model.
S340, selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions.
According to the technical scheme provided by the embodiment of the application, whether the initial questions can form semantic similar sentence pairs with the user questions or not is judged by utilizing a pre-trained semantic similarity model according to the user questions and the text content information and sentence structure information of each initial question, the initial questions which can form the semantic similar sentence pairs with the user questions are determined to be candidate questions, the user questions and the initial questions are matched in semantic characteristics, the initial questions which are successfully matched are taken as candidate questions, the re-screening of the initial questions is achieved, and the searching range of user question answers is further shortened. Compared with the initial problem, the obtained candidate problem has higher similarity with the user problem in the word, is closer to the user problem in sentence meaning, and improves the response accuracy of the question-answering system.
Example IV
Fig. 4 is a flowchart of yet another question-answering processing method provided in the fourth embodiment of the present application. The present embodiment is further optimized on the basis of the above embodiment. Specifically, according to third feature data between the user question and the candidate question, selecting a target question from the candidate questions, and taking an answer associated with the target question as an answer of the user question, including: splicing third characteristic data of the user problem and the candidate problem to be used as input of a neural network model; and determining target questions in the candidate questions according to the number of the sub-questions of the user questions and the categories of the sub-questions output by the neural network model.
As shown in fig. 4, the question-answering processing method includes:
s410, performing first feature matching on the user problem and the pre-stored problems in the pre-stored question-answering pair, and taking at least two pre-stored problems successfully matched as initial problems.
S420, performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem.
And S430, splicing third characteristic data of the user problem and the candidate problem, and using the third characteristic data as input of a neural network model.
The syntactic features, the contextual features and the semantic similarity features are spliced and input into the neural network model as feature vectors. The neural network model is a natural language processing model which is determined by related technicians according to actual conditions and is trained in advance. The neural network model is used for outputting the number of sub-questions of the user question and the category of the sub-questions according to the syntactic features, the contextual features and the semantic similarity features.
S440, determining target questions in the candidate questions according to the number of the sub-questions and the categories of the sub-questions of the user questions output by the neural network model, and taking answers associated with the target questions as answers of the user questions.
The categories of the sub-questions are determined by a technician according to an actual business scene, and exemplary categories of the sub-questions include types of financial accounting, loan, savings and the like. The corresponding answers of the same user questions also have differences in different service scenes, so that in addition to the similarity of the user questions and the candidate questions, factors of the service scenes need to be comprehensively considered. The category of the sub-problem reflects the traffic scenario information.
In addition, in order to ensure that each sub-question in the user question can be answered by a corresponding business scope. In an alternative embodiment, determining a target problem among the candidate problems according to the number of sub-problems of the user problem and the category of the sub-problems output by the neural network model includes: selecting a problem which is consistent with the category of the sub-problem from the candidate problems as a target problem; and determining the number of the target questions according to the number of the sub-questions so as to check whether the sub-questions all have corresponding target questions.
In order for the neural network to output the category of questions and the number of questions including the sub-questions, the neural network is trained using a training data set that includes the category of questions and the number of questions including the sub-questions. The training data set can be selected from a public data set or can be constructed autonomously. A training data set is independently constructed, and illustratively, 5000 samples are randomly extracted from question-answer data interacted by a user-client manager under the condition of human-computer interaction question-answer of a bank, and are handed to a annotator to manually annotate standard questions corresponding to the user questions and common question sets. And calculating the consistency of the labeling result by using the consistency check value, and specifically labeling the question number of the sample and the corresponding standard question category. The consistency check value of the labeling results on the two tasks is larger than 0.7, which indicates that the labeling consistency is higher and the corpus is available, thereby forming a training data set.
The common question set is a pre-stored question and answer pair, can be a publicly used question set, and can also be obtained by arranging questions frequently submitted by a user in financial product marketing according to professional knowledge by service personnel. By way of example, the common question set may be expanded by manually expanding each question into a similar question. If 100 types of standard questions are collected in total, 10 semantically similar expansion questions in each type are taken as pre-stored question-answer pairs, and 1000 standard question-answer pairs are taken as the pre-stored question-answer pairs.
According to the technical scheme provided by the embodiment of the application, the user problem and the third characteristic data of the candidate problem are spliced to be used as the input of the neural network model. And determining a target question in the candidate questions according to the number of the sub-questions and the categories of the sub-questions of the user questions output by the neural network model, and taking the answers associated with the target question as the answers of the user questions. According to the technical scheme provided by the application, the similarity of the user questions and the candidate questions and the factors of the service scene are comprehensively considered, so that each sub-question in the user questions can be answered in a corresponding service range, and the answer accuracy of the question-answering system and the intelligence of the question-answering system are improved.
Example five
Fig. 5 is a question-answering processing device provided in a fifth embodiment of the present application, where the embodiment is applicable to a case where a human-computer interaction question-answering system receives a user question and feeds back a matched answer to the user aiming at the user question. The apparatus may be implemented in software and/or hardware and may be integrated in an electronic device such as a smart terminal.
As shown in fig. 5, the apparatus may include:
the initial question determining module 510 is configured to perform a first feature matching on the user question and a pre-stored question in the pre-stored question-answering pair, and take at least two pre-stored questions successfully matched as initial questions;
the candidate problem determining module 520 is configured to perform a second feature matching on the user problem and the initial problem, and take the initial problem that is successfully matched as a candidate problem;
a target question determination module 530, configured to select a target question from the candidate questions according to third feature data between the user question and the candidate questions, and take an answer associated with the target question as an answer of the user question; wherein the first feature is different from the second feature.
According to the technical scheme provided by the embodiment of the application, the user questions are matched with the questions in the pre-stored question-answering pair, the answers of the user questions are determined in the known question answers, the user questions are subjected to characteristic matching of different layers and different types for a plurality of times with the pre-stored questions in the pre-stored question-answering pair, the searching range of the answers of the user questions is gradually reduced, the pre-stored questions with the highest matching degree with the user questions are determined as target questions, the answers associated with the target questions are used as the answers of the user questions, the answers of each sub-question in the user questions are ensured to be corresponding through the characteristic matching of multiple layers and multiple angles, the answer accuracy of a question-answering system and the intelligence of the question-answering system are improved, and further the user experience is improved.
Optionally, the first feature refers to a frequency of occurrence of a language unit constituting the user question in the pre-stored question, where the language unit includes words or terms; the second feature is a semantic feature; the third characteristic data includes: syntactic features, contextual features, and semantic similarity features.
Optionally, the initial problem determination module 510 includes:
the language unit segmentation module is used for segmenting the user problem into independent language units;
the first similarity score calculation sub-module is used for calculating a first similarity score of the user problem and the pre-stored problem according to the occurrence frequency of each pre-stored problem in the pre-stored question-answer pair of each language unit;
an initial question determination sub-module, configured to select at least two pre-stored questions with the first similarity score being greater than a preset similarity threshold as an initial question.
Optionally, the first similarity score calculation sub-module is specifically configured to:
calculating a first similarity score of the user question and the pre-stored question according to the following formula:
wherein t is i Representing language units in the user question faq j Representing the pre-stored problem, f (t i ,faq j ) Representing t i Faq on pre-stored problem j Frequency of occurrence k 1 And b is a first adjustment coefficient and a second adjustment coefficient, w i Represents the correlation weight, avgFAQ represents the average word length of the pre-stored questions, N represents the number of the pre-stored question-answer pairs, N (t) i ) The representation includes t i Faq of the pre-stored problem j Number, p i Weights are used to represent t i Importance level.
Optionally, the matching the user problem with the initial problem in the second feature, and taking the initial problem successfully matched as the candidate problem includes:
judging whether the initial questions and the user questions form semantic similar sentence pairs or not by utilizing a semantic similarity model according to the user questions, the text content information and sentence structure information of each initial question;
if so, the initial question is determined to be a candidate question.
Optionally, the apparatus further includes: the semantic similarity model training module is used for training the semantic similarity model before judging whether the initial problems and the user problems form semantic similarity sentence pairs or not by utilizing the semantic similarity model according to the user problems and the text content information and sentence structure information of each initial problem;
the semantic similarity model training module comprises: the label data determining submodule is used for determining label data of training sample sentence pairs by utilizing a pre-trained semantic similar sentence pair judging model; wherein the tag data includes: semantic similarity sentences score classification attributes and sentence-to-semantic similarity;
The feature data determining submodule is used for extracting the text content information and the sentence structure information of the training sample sentence pairs as feature data;
the semantic similarity model training sub-module is used for training the semantic similarity model by taking the feature data and the tag data as training data so that the semantic similarity model outputs semantic similarity sentences to classify attribute and sentence-to-semantic similarity scores.
Optionally, the apparatus further includes: the semantic similar sentence pair judgment model training sample construction module is specifically used for constructing a training sample of the semantic similar sentence pair judgment model;
the semantic similar sentence pair judgment model training sample construction module comprises: the positive sample sentence pair constructing sub-module is used for carrying out the first feature matching on the two sample problems in the training sample sentence pair, and if the matching is successful, the training sample sentence pair is used as a positive sample sentence pair;
and the negative example sample sentence pair construction submodule is used for taking the training sample sentence pair as a negative example sample sentence pair if the matching fails.
Optionally, the device further comprises a syntax feature determining module, specifically configured to determine a syntax feature;
The syntax feature determination module includes: the syntax structure information determining submodule is used for respectively analyzing the user problem and the candidate problem by utilizing dependency syntax to obtain syntax structure information; wherein the syntax structure information is information associated with semantics;
a sentence component information and sentence component combination relation extracting sub-module for respectively extracting sentence component information and sentence component combination relation of the user question and the candidate question;
and the syntax feature determining sub-module is used for vectorizing the syntax structure information according to the sentence components and the number of the sentence component combination relations respectively, and splicing vectorized results to be used as syntax features.
Optionally, the apparatus further includes: the context feature determining module is specifically used for determining the context feature;
a contextual feature determination module, comprising: a context semantic information extraction sub-module for respectively extracting context semantic information of the user problem and the candidate problem;
and the semantic features vectorize the context semantic information respectively, and the vectorized results are spliced to be used as the semantic features.
Optionally, the semantic similarity feature is the sentence-to-semantic similarity score output by the semantic similarity model.
Optionally, the objective problem determination module 530 includes:
a third feature data splicing sub-module, configured to splice third feature data of the user problem and the candidate problem, as input of a neural network model;
and the target problem determining sub-module is used for determining target problems in the candidate problems according to the number of sub-problems of the user problems and the category of the sub-problems, which are output by the neural network model.
Optionally, the objective problem determination submodule includes:
a target question determination first unit, configured to select, as a target question, a question that is consistent with the category of the sub-question from among the candidate questions;
and the target problem determining second unit is used for determining the number of the target problems according to the number of the sub-problems so as to check whether the sub-problems all have the corresponding target problems.
The question-answering processing device provided by the embodiment of the invention can execute the question-answering processing method provided by any embodiment of the invention, and has the corresponding functional modules and beneficial effects of executing the question-answering processing method.
Example six
A sixth embodiment of the present application also provides a storage medium containing computer-executable instructions, which when executed by a computer processor, are configured to perform a question-answering processing method, the method comprising:
Performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions;
performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem;
selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions; wherein the first feature is different from the second feature.
Storage media refers to any of various types of memory electronic devices or storage electronic devices. The term "storage medium" is intended to include: mounting media such as CD-ROM, floppy disk or tape devices; computer system memory or random access memory such as DRAM, DDR RAM, SRAM, EDO RAM, lanbas (Rambus) RAM, etc.; nonvolatile memory such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. The storage medium may also include other types of memory or combinations thereof. In addition, the storage medium may be located in a computer system in which the program is executed, or may be located in a different second computer system connected to the computer system through a network (such as the internet). The second computer system may provide program instructions to the computer for execution. The term "storage medium" may include two or more storage media that may reside in different unknowns (e.g., in different computer systems connected by a network). The storage medium may store program instructions (e.g., embodied as a computer program) executable by one or more processors.
Of course, the storage medium containing the computer executable instructions provided in the embodiments of the present application is not limited to the question-answer processing operation described above, and may also perform the relevant operations in the question-answer processing method provided in any embodiment of the present application.
Example seven
The seventh embodiment of the present application provides an electronic device, in which the question-answering processing apparatus provided in the embodiments of the present application may be integrated, where the electronic device may be configured in a system, or may be a device that performs some or all of the functions in the system. Fig. 6 is a schematic structural diagram of an electronic device according to a seventh embodiment of the present application. As shown in fig. 6, the present embodiment provides an electronic device 600, which includes: one or more processors 620; a storage device 610, configured to store one or more programs, where the one or more programs are executed by the one or more processors 620, cause the one or more processors 620 to implement a question-answering method provided by an embodiment of the present application, the method including:
performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions;
Performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem;
selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions; wherein the first feature is different from the second feature.
Of course, those skilled in the art will appreciate that the processor 620 may also implement the solution of the question-answer processing method provided in any embodiment of the present application.
The electronic device 600 shown in fig. 6 is merely an example, and should not be construed as limiting the functionality and scope of use of embodiments of the present application.
As shown in fig. 6, the electronic device 600 includes a processor 620, a storage device 610, an input device 630, and an output device 640; the number of processors 620 in the electronic device may be one or more, one processor 620 being taken as an example in fig. 6; the processor 620, the storage 610, the input 630, and the output 640 in the electronic device may be connected by a bus or other means, as exemplified in fig. 6 by a bus 650.
The storage device 610 is a computer readable storage medium, and may be used to store a software program, a computer executable program, and a module unit, such as program instructions corresponding to the question-answer processing method in the embodiment of the present application.
The storage device 610 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, at least one application program required for functions; the storage data area may store data created according to the use of the terminal, etc. In addition, the storage 610 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid-state storage device. In some examples, the storage device 610 may further include memory remotely located with respect to the processor 620, which may be connected via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 630 may be used to receive input numeric, character information, or voice information, and to generate key signal inputs related to user settings and function control of the electronic device. The output device 640 may include an electronic device such as a display screen, a speaker, etc.
The question-answering processing device, the medium and the electronic equipment provided in the above embodiments can execute the question-answering processing method provided in any embodiment of the application, and have the corresponding functional modules and beneficial effects of executing the method. Technical details not described in detail in the above embodiments may be found in the question-answer processing method provided in any embodiment of the present application.
Note that the above is only a preferred embodiment of the present application and the technical principle applied. Those skilled in the art will appreciate that the present application is not limited to the particular embodiments described herein, but is capable of numerous obvious changes, rearrangements and substitutions as will now become apparent to those skilled in the art without departing from the scope of the present application. Therefore, while the present application has been described in connection with the above embodiments, the present application is not limited to the above embodiments, but may include many other equivalent embodiments without departing from the spirit of the present application, the scope of which is defined by the scope of the appended claims.
Claims (13)
1. A question-answering processing method, characterized in that the method comprises:
performing first feature matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions;
performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem;
selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions; wherein the first feature is different from the second feature; the third characteristic data includes: syntactic, contextual, and semantic similarity features;
Selecting a target question from the candidate questions according to third characteristic data between the user questions and the candidate questions, and taking an answer associated with the target question as an answer of the user questions, wherein the method comprises the following steps: splicing third characteristic data of the user problem and the candidate problem to be used as input of a neural network model; determining target questions in the candidate questions according to the number of the sub-questions of the user questions and the categories of the sub-questions output by the neural network model;
determining a target problem in the candidate problems according to the number of sub-problems of the user problems and the categories of the sub-problems, which are output by the neural network model, wherein the determining comprises the following steps: selecting a problem which is consistent with the category of the sub-problem from the candidate problems as a target problem; and determining the number of the target questions according to the number of the sub-questions so as to check whether the sub-questions all have corresponding target questions.
2. The method according to claim 1, wherein the first characteristic refers to a frequency of occurrence in the pre-stored questions of a language unit constituting the user questions, the language unit including words or phrases; the second feature is a semantic feature.
3. The method of claim 2, wherein the first feature matching the user question with the pre-stored questions in the pre-stored question-and-answer pair, and wherein the at least two pre-stored questions that are successfully matched are used as the initial questions, comprises:
dividing the user problem into independent language units;
calculating a first similarity score of the user problem and the pre-stored problem according to the occurrence frequency of each pre-stored problem in the pre-stored question-answering pair of each language unit;
and selecting at least two pre-stored questions with the first similarity score larger than a preset similarity threshold as initial questions.
4. A method according to claim 3, wherein said calculating a first similarity score for said user question and said pre-stored questions based on the frequency of occurrence of each pre-stored question in said pre-stored question-answer pair by each language unit comprises:
calculating a first similarity score of the user question and the pre-stored question according to the following formula:
wherein t is i Representing language units in the user question faq j Representing the pre-stored problem, f (t i ,faq j ) Representing t i Faq on pre-stored problem j Frequency of occurrence k 1 And b is a first adjustment coefficient and a second adjustment coefficient, w i Represents the correlation weight, avgFAQ represents the average word length of the pre-stored questions, N represents the number of the pre-stored question-answer pairs, N (t) i ) The representation includes t i Faq of the pre-stored problem j Number, p i Is used as a weight to represent t i Importance level.
5. The method of claim 2, wherein the second feature matching the user question with the initial question and taking the initial question that was successfully matched as the candidate question comprises:
judging whether the initial questions and the user questions form semantic similar sentence pairs or not by utilizing a semantic similarity model according to the user questions, the text content information and sentence structure information of each initial question;
if so, the initial question is determined to be a candidate question.
6. The method of claim 5, further comprising the training process of the semantic similarity model before determining whether an initial question can form a semantic similarity sentence pair with the user question based on the text content information and sentence structure information of the user question and each initial question:
determining the label data of training sample sentence pairs by utilizing a pre-trained semantic similar sentence pair judging model; wherein the tag data includes: semantic similarity sentences score classification attributes and sentence-to-semantic similarity;
Extracting the text content information and the sentence structure information of the training sample sentence pair as feature data;
and training the semantic similarity model by using the feature data and the tag data as training data so as to enable the semantic similarity model to output semantic similarity scores of semantic similarity sentences and classification attributes and sentence-to-semantic similarity scores.
7. The method of claim 6, wherein the process of constructing the training sample of the semantic similar sentence pair judgment model is as follows:
performing the first feature matching on the two sample problems in the training sample sentence pair, and taking the training sample sentence pair as a positive sample sentence pair if the matching is successful;
and if the matching fails, taking the training sample sentence pair as a negative example sample sentence pair.
8. The method of claim 2, wherein the method further comprises a process of determining a syntactic characteristic:
analyzing the user problem and the candidate problem by utilizing dependency syntax respectively to obtain syntax structure information; wherein the syntax structure information is information associated with semantics;
extracting sentence component information and sentence component combination relations of the user questions and the candidate questions respectively;
And vectorizing the syntax structure information according to the number of the sentence components and the sentence component combination relations, and splicing vectorized results to serve as syntax features.
9. The method of claim 2, wherein the method further comprises a context feature determination process:
extracting context semantic information of the user problem and the candidate problem respectively;
and vectorizing the context semantic information respectively, and splicing vectorization results to serve as the semantic features.
10. The method of claim 6, wherein the semantic similarity feature is the sentence-to-semantic similarity score output by the semantic similarity model.
11. A question-answering apparatus, the apparatus comprising:
the initial question determining module is used for performing first characteristic matching on the user questions and the pre-stored questions in the pre-stored question-answering pair, and taking at least two pre-stored questions successfully matched as initial questions;
the candidate problem determining module is used for performing second feature matching on the user problem and the initial problem, and taking the initial problem successfully matched as a candidate problem;
a target question determining module, configured to select a target question from the candidate questions according to third feature data between the user question and the candidate questions, and take an answer associated with the target question as an answer of the user question; wherein the first feature is different from the second feature; the third characteristic data includes: syntactic, contextual, and semantic similarity features;
Wherein, the target problem determination module includes: a third feature data splicing sub-module, configured to splice third feature data of the user problem and the candidate problem, as input of a neural network model; a target problem determining sub-module, configured to determine a target problem among the candidate problems according to the number of sub-problems of the user problem and the category of the sub-problems output by the neural network model;
a targeting problem determination sub-module comprising: a target question determination first unit, configured to select, as a target question, a question that is consistent with the category of the sub-question from among the candidate questions; and the target problem determining second unit is used for determining the number of the target problems according to the number of the sub-problems so as to check whether the sub-problems all have the corresponding target problems.
12. A computer-readable storage medium, on which a computer program is stored, which program, when executed by a processor, implements a question-answering method according to any one of claims 1-10.
13. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable by the processor, wherein the processor implements the question-answering method according to any one of claims 1-10 when the computer program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110349133.XA CN112989001B (en) | 2021-03-31 | 2021-03-31 | Question and answer processing method and device, medium and electronic equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110349133.XA CN112989001B (en) | 2021-03-31 | 2021-03-31 | Question and answer processing method and device, medium and electronic equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112989001A CN112989001A (en) | 2021-06-18 |
| CN112989001B true CN112989001B (en) | 2023-05-26 |
Family
ID=76338725
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110349133.XA Active CN112989001B (en) | 2021-03-31 | 2021-03-31 | Question and answer processing method and device, medium and electronic equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112989001B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116204726B (en) * | 2023-04-28 | 2023-07-25 | 杭州海康威视数字技术股份有限公司 | Data processing method, device and equipment based on multi-mode model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108920654A (en) * | 2018-06-29 | 2018-11-30 | 泰康保险集团股份有限公司 | A kind of matched method and apparatus of question and answer text semantic |
| CN111858859A (en) * | 2019-04-01 | 2020-10-30 | 北京百度网讯科技有限公司 | Automatic question-answering processing method, device, computer equipment and storage medium |
-
2021
- 2021-03-31 CN CN202110349133.XA patent/CN112989001B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108920654A (en) * | 2018-06-29 | 2018-11-30 | 泰康保险集团股份有限公司 | A kind of matched method and apparatus of question and answer text semantic |
| CN111858859A (en) * | 2019-04-01 | 2020-10-30 | 北京百度网讯科技有限公司 | Automatic question-answering processing method, device, computer equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112989001A (en) | 2021-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111353310B (en) | Named entity identification method and device based on artificial intelligence and electronic equipment | |
| Yu et al. | Deep learning for answer sentence selection | |
| WO2019153737A1 (en) | Comment assessing method, device, equipment and storage medium | |
| CN110110062B (en) | Machine intelligent question and answer method and device and electronic equipment | |
| KR102288249B1 (en) | Information processing method, terminal, and computer storage medium | |
| US8204751B1 (en) | Relevance recognition for a human machine dialog system contextual question answering based on a normalization of the length of the user input | |
| US20230069935A1 (en) | Dialog system answering method based on sentence paraphrase recognition | |
| CN109472022B (en) | New word recognition method based on machine learning and terminal equipment | |
| CN110678882A (en) | Selecting answer spans from electronic documents using machine learning | |
| CN116304748B (en) | Text similarity calculation method, system, equipment and medium | |
| CN112036705A (en) | Quality inspection result data acquisition method, device and equipment | |
| CN111782793A (en) | Intelligent customer service processing method, system and equipment | |
| CN114003682A (en) | Text classification method, device, equipment and storage medium | |
| CN113779227A (en) | Case fact extraction method, system, device and medium | |
| Arbaaeen et al. | Natural language processing based question answering techniques: A survey | |
| CN112989001B (en) | Question and answer processing method and device, medium and electronic equipment | |
| US20220245179A1 (en) | Semantic phrasal similarity | |
| CN115905187B (en) | Intelligent proposition system oriented to cloud computing engineering technician authentication | |
| Otani et al. | Large-scale acquisition of commonsense knowledge via a quiz game on a dialogue system | |
| CN113742445B (en) | Text recognition sample obtaining method and device and text recognition method and device | |
| CN116186219A (en) | Man-machine dialogue interaction method, system and storage medium | |
| CN116186220A (en) | Information retrieval method, question and answer processing method, information retrieval device and system | |
| Xie et al. | Focusing attention network for answer ranking | |
| Curini et al. | Sentiment Analysis and Social Media | |
| Rosander et al. | Email Classification with Machine Learning and Word Embeddings for Improved Customer Support |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |