Semantic tendency analysis method and system based on multi-round questions and answers

Abstract

The invention relates to the technical field of natural language processing, and discloses a semantic tendency analysis method and system based on multi-round questions and answers, and the system comprises a data collection module and a semantic analysis module. According to the semantic tendency analysis method and system based on multiple rounds of questions and answers, a data acquisition module is in network connection with a big data platform and a database, historical dialogue data of all users and knowledge information of all clusters are obtained and classified to form a data set, and a semantic analysis module selects different text processing methods to analyze the semantic tendency. Syntactic analysis is carried out by using a natural language processing technology, a corresponding word segmentation data set is generated, knowledge information in each cluster in the knowledge data set is retrieved, a corresponding matching data set is generated, whether secondary inquiry is initiatively initiated or not is judged, the capability of flexibly coping with multi-round question and answer analysis is high, and when a user replies the secondary inquiry, the user experience is improved. The answer satisfaction degree is obtained through the question and answer data set, the semantic tendency analysis accuracy is judged, the corresponding judgment result is output, and the semantic tendency analysis accuracy is high.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a semantic tendency analysis method and system based on multi-round question and answer. Background Art

[0002] The semantic tendency analysis system based on multi-round question-answering uses natural language processing, deep learning, knowledge graphs and other technologies to understand and analyze user intentions, emotional tendencies and contextual information through multi-round dialogues, thereby providing an intelligent system with more accurate answers or solutions. Natural language processing technology enables the analysis and understanding of user questions and converts them into a machine-readable format. Knowledge graph technology provides a rich knowledge base and entity links, providing strong knowledge support for the semantic tendency analysis system. Deep learning technology provides core technical support for the semantic tendency analysis system by parsing and understanding natural language text. The semantic tendency analysis system can not only help individuals and enterprises better understand and respond to changes in the world around them, but also provide in-depth insights and value in multiple fields.

[0003] At present, traditional semantic tendency analysis systems find it difficult to flexibly respond to diverse conversation content when processing multiple rounds of long text conversations. The results of semantic tendency analysis for users are often vague, and when the keywords of user questions need to be confirmed again, it is also difficult to clearly express the content of the inquiry. The reply content of the search results is limited by the reserve capacity of the database, and it is difficult to provide more intelligent reply content. Summary of the invention

[0004] 1. Technical issues to be resolved In view of the shortcomings of the prior art, the present invention provides a semantic tendency analysis method and system based on multi-round question and answer, which has the advantages of strong ability to flexibly cope with multi-round question and answer analysis and high accuracy of semantic tendency analysis. It solves the problem that traditional semantic tendency analysis systems are difficult to flexibly cope with long text conversations and have low accuracy of semantic tendency analysis.

[0005] (II) Technical solution To achieve the above object, the present invention provides the following technical solution: a semantic tendency analysis method based on multi-round question and answer, comprising the following steps: Step 1: Obtain all users’ historical conversation data and all clustered knowledge information through the big data platform and database, and classify them into question-answering datasets and knowledge datasets; Step 2: Select different text processing methods based on the question-answering dataset, and then use natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation dataset. ; Step 3: According to the word segmentation data group , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , determine whether to initiate a secondary inquiry ; Step 4: When the user responds to the second inquiry When , we use the question-answering dataset to obtain the answer satisfaction , judge the accuracy of semantic tendency analysis and output the corresponding judgment result.

[0006] Preferably, in step 1, the expression of the question-answering dataset is , to The first to the The historical conversation data of each user includes historical questions, answer content, and answer satisfaction. , Represents the total number of conversation rounds for a single user in a multi-round question-answering process.

[0007] Preferably, in step 1, the expression of the knowledge data set is , to The first to the The knowledge information of clusters, The number representing a single cluster.

[0008] Preferably, in step 2, the word segmentation data set The calculation process is as follows: Extract the first The historical conversation data of users is marked as ; according to , will The historical questions entered by users in the first round are marked as , Statistics The number of words in the historical questions entered by users in the first round and marked as ; according to , select a text processing method, and then use natural language processing technology to Perform syntactic analysis on the text content in like , indicating that the number of words in the first round of history questions is less than 50, and first-level word segmentation is required; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into a set of content words according to the part of speech and function word set ; like , indicating that the number of words in the first round of history questions is greater than or equal to 50, and first-level word segmentation and second-level word segmentation are required; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into subject sets according to grammatical functions , predicate set , object set , attributive set , adverbial set , preposition set , adjective set , the central word set and independent word sets .

[0009] Preferably, in step 3, matching data set The calculation process is as follows: Extract the knowledge from the dataset The knowledge information of clusters is marked as ; ; In the formula, Represents a matching data set, Represents the content word set and the The intersection of clustered knowledge information is used for retrieval matching after the first-level word segmentation processing. The retrieval result is the answer content corresponding to the first round of historical questions. Indicates the central word set and the The intersection of clustered knowledge information is used for retrieval matching after secondary word segmentation processing. Represents a set of independent words and The intersection of clustered knowledge information is used for retrieval matching after secondary word segmentation processing. Represents the main sentence in the first round of history questions, Indicates the sentence stem and the first round of history questions The intersection of the clustered knowledge information is used for retrieval matching after the secondary word segmentation processing, and the retrieval result is the answer content corresponding to the first round of historical questions.

[0010] Preferably, in step 3, if the corresponding matching data set cannot be retrieved and generated , indicating that there are limitations in the database and a second query needs to be initiated , if the matching data set If the intersection of any set and the clustering knowledge information is an empty set, it means that the answer is incomplete and a second inquiry needs to be initiated. .

[0011] Preferably, in step three, the second inquiry The calculation process is as follows: If the corresponding matching data set cannot be retrieved and generated , proactive secondary inquiry These are the responses to the first round of historical questions. ; In the formula, Indicates a second inquiry. Indicates the key words in the first round of history questions; If the matching data set The intersection of any set and the clustering knowledge information is an empty set. While searching for results, a secondary query is initiated It is also considered as one of the responses to the first round of historical questions. ; In the formula, Indicates a second inquiry. Represents the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets. Represents the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets. Represents the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets.

[0012] Preferably, in step 3, if the matching data set When there is common data in the intersection of each set and cluster knowledge information, it means that the answer is complete, the first round of question and answer has ended, and there is no need to actively initiate a second inquiry. .

[0013] Preferably, in step 4, if the response satisfaction When the percentage is less than 60%, it means that the user is not satisfied with the answer content and the accuracy of semantic tendency analysis is low. Return to step 3 until the answer satisfaction is reached. Greater than or equal to 60%.

[0014] A semantic tendency analysis system based on multi-round question and answer, including a data collection module and a semantic analysis module; The data collection module is composed of a historical question-and-answer unit and a knowledge storage unit. The historical question-and-answer unit collects a question-and-answer data set through a network connection to a big data platform. The question-and-answer data set includes historical conversation data of all users. The knowledge storage unit collects a knowledge data set through a network connection to a database. The knowledge data set includes knowledge information of all clusters. The data collection module transmits the question-and-answer data set and the knowledge data set to the semantic analysis module through the network. The semantic analysis module consists of a syntactic analysis unit, a retrieval matching unit, and an evaluation management unit. The syntactic analysis unit selects different text processing methods according to the question-answer data set, and then uses natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation data set. The search matching unit is based on the word segmentation data group , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , determine whether to initiate a secondary inquiry The evaluation management unit obtains the answer satisfaction through the question and answer dataset , judge the accuracy of semantic tendency analysis and output the corresponding judgment result.

[0015] Compared with the prior art, the present invention provides a semantic tendency analysis method and system based on multi-round question-answering, which has the following beneficial effects: 1. The present invention connects the big data platform and the database through the data acquisition module network to obtain the historical conversation data of all users and the knowledge information of all clusters, and classifies them into question and answer data sets and knowledge data sets. The semantic analysis module selects different text processing methods according to the question and answer data sets. The number of words in the first round of historical questions is greater than or equal to 50 words, and the question content is a long paragraph of text. It is necessary to perform primary and secondary word segmentation processing in combination with the context, deeply analyze the corresponding retrieval direction, and then use natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation data set. , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , determine whether to initiate a secondary inquiry If the corresponding matching data set cannot be retrieved and generated , indicating that there are limitations in the database or that the user input is incorrect, and a second query needs to be initiated , if the matching data set If the intersection of any set and the clustering knowledge information is an empty set, it means that the answer is incomplete and a second inquiry needs to be initiated. , and has strong ability to flexibly respond to multiple rounds of question and answer analysis.

[0016] 2. The present invention obtains answer satisfaction through the question-answering dataset through the semantic analysis module , judge the accuracy of semantic tendency analysis, if the answer is satisfactory When the number is less than 60%, it means that the user is not satisfied with the answer content and the accuracy of semantic tendency analysis is low. Return to step 3 and search again until the answer is satisfactory. Greater than or equal to 60%, the semantic tendency analysis has high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a step diagram of the method of the present invention; Figure 2 It is a schematic diagram of the system flow of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Because traditional semantic tendency analysis systems are difficult to flexibly respond to diverse conversation content when processing multiple rounds of long text conversations, the semantic tendency analysis results for users are often vague. When the keywords of the user's question need to be confirmed again, it is also difficult to clearly express the query content. The reply content of the search results is limited by the reserve capacity of the database, and it is difficult to provide more intelligent reply content. Therefore, a semantic tendency analysis method and system based on multi-round question and answer are provided. Please refer to Figure 1 ,The semantic tendency analysis method based on multi-round question and answer includes the following steps: Step 1: Obtain all users’ historical conversation data and all clustered knowledge information through the big data platform and database, and classify them into question-answering datasets and knowledge datasets; The expression of the question-answering dataset is , to The first to the The historical conversation data of each user includes historical questions, answer content, and answer satisfaction. , Indicates the total number of conversation rounds for a single user in a multi-round question-and-answer process. In actual use, the big data platform will actively ask users about their satisfaction with the answer content at the end of each round of question-and-answer. , with 10%, 20%, 30%, ..., 100% options, which will help to optimize the semantic tendency analysis algorithm in a targeted manner later; The expression of the knowledge dataset is , to The first to the The knowledge information of clusters, The number of a single cluster. The knowledge information in the same cluster has a high degree of similarity, while the information characteristics between different clusters are quite different, which provides a clear direction for the system to analyze the user's semantic tendency. Step 2: Select different text processing methods based on the question-answering dataset, and then use natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation dataset. , the calculation process is as follows: Extract the first The historical conversation data of users is marked as ; according to , will The historical questions entered by users in the first round are marked as , Statistics The number of words in the historical questions entered by users in the first round and marked as ; according to , select a text processing method, and then use natural language processing technology to Perform syntactic analysis on the text content in like , which means that the number of words in the first round of history questions is less than 50, the question content is relatively brief, the corresponding search direction is very clear, and first-level word segmentation processing is required; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into a set of content words according to the part of speech and function word set ; like , which means that the number of words in the first round of history questions is greater than or equal to 50, and the question content is a long paragraph of text, which needs to be processed by first-level and second-level word segmentation in combination with the context, and the corresponding search direction needs to be analyzed in depth; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into subject sets according to grammatical functions , predicate set , object set , attributive set , adverbial set , preposition set , adjective set , the central word set and independent word sets ,The text content after word segmentation removes irrelevant information and noise, ensuring the quality and availability of key data, and significantly improving the accuracy of subsequent semantic tendency analysis; Step 3: According to the word segmentation data group , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , the calculation process is as follows: Extract the knowledge from the dataset The knowledge information of clusters is marked as ; ; In the formula, Represents a matching data set, Represents the content word set and the The intersection of clustered knowledge information is used for retrieval matching after the first-level word segmentation processing. The retrieval result is the answer content corresponding to the first round of historical questions. Indicates the central word set and the The intersection of clustered knowledge information is used for retrieval matching after secondary word segmentation processing. Represents a set of independent words and The intersection of clustered knowledge information is used for retrieval matching after secondary word segmentation processing. Represents the main sentence in the first round of history questions, Indicates the sentence stem and the first round of history questions The intersection of clustered knowledge information is used for retrieval matching after the second-level word segmentation processing. The retrieval result is the answer content corresponding to the first round of historical questions. According to the matching data group , determine whether to initiate a secondary inquiry If the corresponding matching data set cannot be retrieved and generated , indicating that there are limitations in the database or that the user input is incorrect, and a second query needs to be initiated , if the matching data set If the intersection of any set and the clustering knowledge information is an empty set, it means that the answer is incomplete and a second inquiry needs to be initiated. ; Second inquiry The calculation process is as follows: If the corresponding matching data set cannot be retrieved and generated , proactive secondary inquiry These are the responses to the first round of historical questions. ; In the formula, Indicates a second inquiry. Indicates the key words in the first round of history questions; If the matching data set The intersection of any set and the clustering knowledge information is an empty set. While searching for results, a secondary query is initiated It is also considered as one of the responses to the first round of historical questions. ; In the formula, Indicates a second inquiry. Indicates the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets. Indicates the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets. Represents the first The knowledge information of clusters, and Matching data set There is an intersection in any set, and the ability to flexibly deal with multiple rounds of question-answering analysis is strong; If the matching data set When there is common data in the intersection of each set and cluster knowledge information, it means that the answer is complete, the first round of question and answer has ended, and there is no need to actively initiate a second inquiry. ; Step 4: When the user responds to the second inquiry When , we use the question-answering dataset to obtain the answer satisfaction , judge the accuracy of semantic tendency analysis, if the answer is satisfactory When the number is less than 60%, it means that the user is not satisfied with the answer content and the accuracy of semantic tendency analysis is low. Return to step 3 and search again until the answer is satisfactory. Greater than or equal to 60%, the semantic tendency analysis has high accuracy.

[0020] See also Figure 2,A semantic tendency analysis system based on multi-round question and answer, including a data collection module and a semantic analysis module; The data collection module consists of a historical question-and-answer unit and a knowledge storage unit. The historical question-and-answer unit collects question-and-answer data sets through a network connection to the big data platform. The question-and-answer data sets include all users’ historical conversation data. The knowledge storage unit collects knowledge data sets through a network connection to the database. The knowledge data sets include all clustered knowledge information. The data collection module transmits the question-and-answer data sets and knowledge data sets to the semantic analysis module through the network. The semantic analysis module consists of a syntactic analysis unit, a retrieval matching unit, and an evaluation management unit. The syntactic analysis unit selects different text processing methods based on the question-answering data set, and then uses natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation data set. , the search matching unit is based on the word segmentation data group , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , determine whether to initiate a secondary inquiry , and has strong ability to flexibly respond to multiple rounds of question and answer analysis. The evaluation management unit obtains answer satisfaction through the question and answer data set , judge the accuracy of semantic tendency analysis, and output the corresponding judgment result, the accuracy of semantic tendency analysis is high.

[0021] Example 1: In this experiment, user A is selected as the experimental subject. The historical question entered by user A in the first round is "What does cos mean?" The word segmentation data set corresponding to this question is The calculation process is as follows: The word count for the first round of history questions is 7. , indicating that the number of words in the first round of history questions is less than 50, and first-level word segmentation is required; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into a set of content words according to the part of speech and function word set , and with " " is further searched for matches in the word segmentation form. As a mathematical term, it represents the cosine function, and the corresponding cluster is the trigonometric function; Example 2: In this experiment, user B is selected as the experimental subject. The historical question entered by user B in the first round is "the degree of a triangle". However, during the semantic tendency analysis, user B did not specify which angle of the triangle "degree" refers to. The corresponding secondary query of this question is The calculation process is as follows: ; In the formula, It indicates secondary inquiry. There is an intersection between the knowledge information of right triangle and triangle cluster, there is an intersection between the knowledge information of acute triangle and triangle cluster, and there is an intersection between the knowledge information of obtuse triangle and triangle cluster. It is necessary to further inquire whether user B's semantic preference is right triangle, acute triangle, or obtuse triangle.

[0022] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A semantic tendency analysis method based on multi-round question-answering, characterized in that: The following steps are involved: Step 1: Obtain all users’ historical conversation data and all clustered knowledge information through the big data platform and database, and classify them into question-answering datasets and knowledge datasets; Step 2: Select different text processing methods based on the question-answering dataset, and then use natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation dataset. ; Step 3: According to the word segmentation data group , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , determine whether to initiate a secondary inquiry ; Step 4: When the user responds to the second inquiry When , we use the question-answering dataset to obtain the answer satisfaction , judge the accuracy of semantic tendency analysis and output the corresponding judgment result.

2. The semantic tendency analysis method based on multi-round question-answering according to claim 1 is characterized in that: In step 1, the expression of the question-answering dataset is: , to The first to the The historical conversation data of each user includes historical questions, answer content, and answer satisfaction. , Represents the total number of conversation rounds for a single user in a multi-round question-answering process.

3. The semantic tendency analysis method based on multi-round question and answer according to claim 2 is characterized in that: In step 1, the expression of the knowledge data set is: , to The first to the The knowledge information of clusters, The number representing a single cluster.

4. The semantic tendency analysis method based on multi-round question-answering according to claim 3 is characterized in that: In the step 2, the word segmentation data set The calculation process is as follows: Extract the first The historical conversation data of users is marked as ; according to , will The historical questions entered by users in the first round are marked as , Statistics The number of words in the historical questions entered by users in the first round and marked as ; according to , select a text processing method, and then use natural language processing technology to Perform syntactic analysis on the text content in like , indicating that the number of words in the first round of history questions is less than 50, and first-level word segmentation is required; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into a set of content words according to the part of speech and function word set ; like , indicating that the number of words in the first round of history questions is greater than or equal to 50, and first-level word segmentation and second-level word segmentation are required; ; In the formula, Represents a word segmentation data group, Indicates that the text content is divided into subject sets according to grammatical functions , predicate set , object set , attributive set , adverbial set , preposition set , adjective set , the central word set and independent word sets .

5. The semantic tendency analysis method based on multi-round question-answering according to claim 4 is characterized in that: In step 3, matching data set The calculation process is as follows: Extract the knowledge from the dataset The knowledge information of clusters is marked as ; ; In the formula, Represents a matching data set, Represents the content word set and the The intersection of clustered knowledge information is used for retrieval matching after the first-level word segmentation processing. The retrieval result is the answer content corresponding to the first round of historical questions. Indicates the central word set and the The intersection of clustered knowledge information is used for retrieval matching after secondary word segmentation processing. Represents a set of independent words and The intersection of clustered knowledge information is used for retrieval matching after secondary word segmentation processing. Represents the main sentence in the first round of history questions, Indicates the sentence stem and the first round of history questions The intersection of the clustered knowledge information is used for retrieval matching after the secondary word segmentation processing, and the retrieval result is the answer content corresponding to the first round of historical questions.

6. The semantic tendency analysis method based on multi-round question-answering according to claim 5 is characterized in that: In step 3, if the corresponding matching data set cannot be retrieved and generated , indicating that there are limitations in the database and a second query needs to be initiated , if the matching data set If the intersection of any set and the clustering knowledge information is an empty set, it means that the answer is incomplete and a second inquiry needs to be initiated. .

7. The semantic tendency analysis method based on multi-round question-answering according to claim 6 is characterized in that: In step 3, the second inquiry The calculation process is as follows: If the corresponding matching data set cannot be retrieved and generated , proactive secondary inquiry These are the responses to the first round of historical questions. ; In the formula, Indicates a second inquiry. Indicates the key words in the first round of history questions; If the matching data set The intersection of any set and the clustering knowledge information is an empty set. While searching for results, a secondary query is initiated It is also considered as one of the responses to the first round of historical questions. ; In the formula, Indicates a second inquiry. Indicates the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets. Indicates the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets. Indicates the first The knowledge information of clusters, and Matching data set There is an intersection between any of the sets.

8. The semantic tendency analysis method based on multi-round question-answering according to claim 7 is characterized in that: In step 3, if the matching data set When there is common data in the intersection of each set and cluster knowledge information, it means that the answer is complete, the first round of question and answer has ended, and there is no need to actively initiate a second inquiry. .

9. The semantic tendency analysis method based on multi-round question-answering according to claim 8, characterized in that: In step 4, if the response is satisfactory When the percentage is less than 60%, it means that the user is not satisfied with the answer content and the accuracy of semantic tendency analysis is low. Return to step 3 until the answer satisfaction is reached. Greater than or equal to 60%.

10. A semantic tendency analysis system based on multi-round question-answering, applied to the semantic tendency analysis method based on multi-round question-answering according to any one of claims 1 to 9, characterized in that: Including data collection module and semantic analysis module; The data collection module is composed of a historical question-and-answer unit and a knowledge storage unit. The historical question-and-answer unit collects a question-and-answer data set through a network connection to a big data platform. The question-and-answer data set includes historical conversation data of all users. The knowledge storage unit collects a knowledge data set through a network connection to a database. The knowledge data set includes knowledge information of all clusters. The data collection module transmits the question-and-answer data set and the knowledge data set to the semantic analysis module through the network. The semantic analysis module consists of a syntactic analysis unit, a retrieval matching unit, and an evaluation management unit. The syntactic analysis unit selects different text processing methods according to the question-answering data set, and then uses natural language processing technology to perform syntactic analysis to generate the corresponding word segmentation data set. The search matching unit is based on the word segmentation data group , retrieve the knowledge information in each cluster in the knowledge data set and generate the corresponding matching data set , determine whether to initiate a secondary inquiry The evaluation management unit obtains the answer satisfaction through the question and answer dataset , judge the accuracy of semantic tendency analysis and output the corresponding judgment result.

Images