Information processing method and device based on semantic understanding and computer device

Abstract

This application relates to a semantic understanding-based information processing method, apparatus, computer device, and storage medium. The method includes: identifying the information category of each piece of information based on semantic understanding technology; associating and storing the corresponding information according to the information category; retrieving the information category from an intelligent knowledge base and extracting the corresponding current information based on the query request information; returning the current information to the terminal; determining the matching degree information between the query request information and the current information based on user feedback; determining the training direction of the semantic training model; training the matching degree information corresponding to each piece of information according to the training direction; and optimizing the information category of each piece of information based on the training results. This method classifies information and implements hierarchical management based on semantic understanding technology, enabling rapid retrieval of information matching the query request information according to information category, and optimizing the information classification based on user feedback to improve data retrieval efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing, in particular to an information processing method and device based on semantic understanding, a computer device and a storage medium. BACKGROUND

[0002] An intelligent knowledge base is a knowledge management system integrating knowledge collection, knowledge update, knowledge recommendation, knowledge sharing and knowledge exchange based on artificial intelligence technology.

[0003] In related technologies, an advisory service management platform established by using an intelligent knowledge base receives query request information from a terminal, searches for corresponding information in the intelligent knowledge base according to the query request information, and returns the information to the corresponding terminal to realize the function of advisory service.

[0004] However, since the intelligent knowledge base lacks classification of knowledge information based on semantic understanding, it is difficult to accurately and efficiently search for knowledge information of the intelligent knowledge base in application scenarios, and service diversion cannot be realized according to information classification, so it is necessary to frequently transfer artificial customer service for processing, and therefore, there are problems of low data search efficiency and low automatic service efficiency. SUMMARY

[0005] Therefore, it is necessary to provide an information processing method and device based on semantic understanding capable of improving data search efficiency, a computer device and a storage medium.

[0006] In a first aspect, the present application provides an information processing method based on semantic understanding, comprising:

[0007] According to the structure, dialogue context and application scenario of each information in a source database, the information category of each information is confirmed based on semantic understanding technology, and corresponding target information is associated and stored in an intelligent knowledge base according to a target information category;

[0008] According to a query request information, a current target information category matching the query request information is searched in the intelligent knowledge base, and corresponding current target information is extracted according to the current target information category;

[0009] The current target information is returned to a terminal corresponding to the query request information, and matching degree information of the query request information and the current target information is determined according to user feedback data of the terminal received for the target information;

[0010] Obtain a semantic training model, determine a training direction of the semantic training model based on target information of a matching degree information anomaly, train matching degree information corresponding to each information of the intelligent knowledge base according to the training direction, obtain a training result, and optimize information categories of each information according to the training result.

[0011] In one of the embodiments, the information categories of each information are determined according to the structure of each information in the source database, the relevance between information, and the application scenario, including:

[0012] According to the structure of information, the information categories of the information as question and answer knowledge or document knowledge are determined;

[0013] According to the dialogue context of information, information applied to the same session is determined as the same information category and the session category;

[0014] According to the application scenario of information, information applied to the same scenario is determined as the same information category and the scenario category.

[0015] In one of the embodiments, the current target information category matching the query request information is retrieved from the intelligent knowledge base according to the query request information, including:

[0016] According to the query request information, a keyword is extracted, and a first target information category matching the keyword is quickly retrieved from the intelligent knowledge base;

[0017] According to the application scenario or subject field involved in the query request information, a second target information category matching the query request information is comprehensively retrieved from the intelligent knowledge base;

[0018] The first target information and the second target information constitute the current target information category.

[0019] In one of the embodiments, the matching degree information of the query request information and the current target information is determined according to the user feedback data of the terminal received for the target information, including:

[0020] If the user feedbacks that the query request information matches successfully, the matching success rate is determined as 1, and the matching success rate is taken as the matching degree information;

[0021] If the user feedbacks that the query request information and the current target information do not match, the query request information is re-searched and re-matched until the user feedbacks that the query request information and the current target information match;

[0022] Record the number of times the query request information is matched and the number of information matched each time. The sum of the number of information matched each time is taken as the total number of matched information. The sum of the number of successfully matched information each time is taken as the total number of successfully matched information. The ratio of the total number of successfully matched information to the total number of matched information is taken as the matching success rate. The matching success rate is taken as the matching degree information.

[0023] In one embodiment, the method further includes:

[0024] In the context of consulting services, based on the query request information, entity information matching the query request information is retrieved from the intelligent knowledge base, and the entity information is returned to the terminal corresponding to the query request information.

[0025] In the context of investigation and handling of business, the entity information in the query request information is verified and checked to obtain result information, and the result information is returned to the terminal corresponding to the query request information.

[0026] In one embodiment, the method further includes:

[0027] The system acquires and analyzes user search history, click behavior, and feedback information to generate user suggestions, which are then returned to the terminal corresponding to the query request information.

[0028] In one embodiment, training the matching degree information corresponding to each piece of information in the intelligent knowledge base according to the training direction includes:

[0029] During the training process, if the matching degree information corresponding to the current information reaches a preset threshold, it is determined that the training of the current information is complete.

[0030] When the number of trained information reaches a preset number, the training process of the semantic training model ends.

[0031] The semantic training model outputs the trained information category and updates the information category corresponding to the target information with abnormal matching degree information to the trained information category.

[0032] Secondly, this application also provides an information processing apparatus based on semantic understanding, comprising:

[0033] The management module is used to determine the information category of each piece of information based on semantic understanding technology, according to the structure of each piece of information in the source database, the dialogue context and the application scenario, and to associate and store the corresponding target information in the intelligent knowledge base according to the target information category.

[0034] The matching module is used to retrieve the current target information category that matches the query request information from the intelligent knowledge base based on the query request information, and extract the corresponding current target information based on the current target information category;

[0035] The entity module is used to return the current target information to the terminal corresponding to the query request information, and determine the matching degree information between the query request information and the current target information based on the user feedback data on the target information received by the terminal.

[0036] The training module is used to acquire a semantic training model, determine the training direction of the semantic training model based on the target information with abnormal matching degree information, train the matching information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtain the training result, and optimize the information category of each piece of information according to the training result.

[0037] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0038] Based on semantic understanding technology, the information category of each piece of information is determined according to the structure, dialogue context and application scenario of each piece of information in the source database, and the corresponding target information is associated and stored in the intelligent knowledge base according to the target information category.

[0039] Based on the query request information, the system retrieves the current target information category that matches the query request information from the intelligent knowledge base, and extracts the corresponding current target information based on the current target information category.

[0040] The current target information is returned to the terminal corresponding to the query request information, and the matching degree information between the query request information and the current target information is determined based on the user feedback data on the target information received by the terminal.

[0041] A semantic training model is obtained. The training direction of the semantic training model is determined based on the target information with abnormal matching degree information. The matching degree information corresponding to each piece of information in the intelligent knowledge base is trained according to the training direction to obtain the training result. The information category of each piece of information is optimized according to the training result.

[0042] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0043] Based on semantic understanding technology, the information category of each piece of information is determined according to the structure, dialogue context and application scenario of each piece of information in the source database, and the corresponding target information is associated and stored in the intelligent knowledge base according to the target information category.

[0044] Based on the query request information, the system retrieves the current target information category that matches the query request information from the intelligent knowledge base, and extracts the corresponding current target information based on the current target information category.

[0045] The current target information is returned to the terminal corresponding to the query request information, and the matching degree information between the query request information and the current target information is determined based on the user feedback data on the target information received by the terminal.

[0046] A semantic training model is obtained. The training direction of the semantic training model is determined based on the target information with abnormal matching degree information. The matching degree information corresponding to each piece of information in the intelligent knowledge base is trained according to the training direction to obtain the training result. The information category of each piece of information is optimized according to the training result.

[0047] The aforementioned information processing method, apparatus, computer equipment, and storage medium based on semantic understanding, utilize semantic understanding technology to determine the information category of each piece of information based on the structure, dialogue context, and application scenario of each piece of information in the source database. The corresponding target information is then associated and stored in an intelligent knowledge base according to the target information category. Based on the query request information, the intelligent knowledge base is used to retrieve the current target information category that matches the query request information, and the corresponding current target information is extracted based on the current target information category. The current target information is returned to the terminal corresponding to the query request information. Based on user feedback data received by the terminal regarding the target information, the matching degree information between the query request information and the current target information is determined. A semantic training model is obtained, and the training direction of the semantic training model is determined based on target information with abnormal matching degree information. The matching degree information corresponding to each piece of information in the intelligent knowledge base is trained according to the training direction to obtain training results. The information category of each piece of information is optimized based on the training results. In this invention, information is classified and hierarchically managed based on semantic understanding technology. Information matching the query request information can be quickly retrieved according to information category, and the classification of information can be further optimized based on user feedback to further improve data retrieval efficiency. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is an application environment diagram of an information processing method based on semantic understanding in one embodiment;

[0050] Figure 2 This is a flowchart illustrating an information processing method based on semantic understanding in one embodiment;

[0051] Figure 3 This is a flowchart illustrating the process of determining the information category of each piece of information based on the structure, dialogue context, and application scenario of each piece of information in the source database, as shown in one embodiment.

[0052] Figure 4 This is a flowchart illustrating the process of retrieving and matching information categories in an intelligent knowledge base according to a query request, as shown in one embodiment.

[0053] Figure 5 This is a flowchart illustrating the process of determining the matching degree information between query request information and information based on user feedback data in one embodiment.

[0054] Figure 6 This is a flowchart illustrating an information processing method based on semantic understanding in yet another embodiment;

[0055] Figure 7 This is a flowchart illustrating an information processing method based on semantic understanding in another embodiment;

[0056] Figure 8 This is a schematic diagram of the process of training the semantic training model to train the matching degree information corresponding to each piece of information in one embodiment;

[0057] Figure 9 This is a schematic diagram of the information flow of an information processing method based on semantic understanding in one embodiment;

[0058] Figure 10 This is a structural block diagram of an information processing device based on semantic understanding in one embodiment;

[0059] Figure 11 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0061] The semantic understanding-based information processing method provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. Server 104, based on semantic understanding technology, determines the information category of each piece of information according to the structure, dialogue context, and application scenario of the information in the source database, and stores the corresponding target information in an intelligent knowledge base according to the target information category. Terminal 102 sends a query request to server 104. Server 104 retrieves the current target information category matching the query request from the intelligent knowledge base, extracts the corresponding current target information according to the current target information category, and returns the current target information to terminal 102. Terminal 102 sends user feedback data based on this target information to server 104. Server 104 determines the matching degree information between the query request and the current target information based on the user feedback data. Server 104 acquires a semantic training model, determines the training direction of the semantic training model based on target information with abnormal matching degree information, trains the matching degree information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtains the training results, and optimizes the information category of each piece of information based on the training results.

[0062] The data storage system can store the data that the server 104 needs to process, and the source database and intelligent knowledge base can be stored in the data storage system. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers.

[0063] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices and portable wearable devices, and the server 104 can be implemented using a standalone server or a server cluster composed of multiple servers.

[0064] In one exemplary embodiment, such as Figure 2 As shown, a semantic understanding-based information processing method is provided, which can be applied to... Figure 1 Taking the server in the example, the explanation includes the following steps S202 to S208. Wherein:

[0065] Step S202: Based on semantic understanding technology, the information category of each piece of information is determined according to the structure, dialogue context and application scenario of each piece of information in the source database, and the corresponding target information is associated and stored in the intelligent knowledge base according to the target information category.

[0066] Semantic understanding technology refers to technologies that can deeply understand the meaning, context, and implicit information of natural language. Information categories refer to the ways in which information content is logically classified and organized. A source database is a database that stores initial data; it can be represented as data to be processed. An intelligent knowledge base is a database used to store, organize, and manage information; it can be represented as a structured management system for knowledge of various topics, domains, or entities.

[0067] For example, based on semantic understanding technology, information is classified into appropriate information categories according to its meaning and semantic relationships based on the attributes of each piece of information in the source database, such as structure, dialogue context, and application scenario. The corresponding information is then associated and stored in an intelligent knowledge base according to different information categories.

[0068] Optionally, in semantic understanding technology, the meaning of a sentence can be understood through grammatical and semantic analysis, that is, by analyzing the structure of the sentence, the relationship between phrases, and the grammatical rules of the sentence; positive, negative, or neutral emotions can be identified by analyzing the sentiment tendency in the text; and implicit information in the text can be inferred by analyzing the context.

[0069] Optionally, the same information may be identified as having one information category or multiple information categories; if the same information is identified as having multiple information categories, then the multiple information categories may be in different thematic ranges or different hierarchical structures.

[0070] Step S204: Based on the query request information, retrieve the current target information category that matches the query request information from the intelligent knowledge base, and extract the corresponding current target information based on the current target information category.

[0071] Among them, query request information refers to the request sent by the user or system to the server to instruct the server to return specific data or information; query request information can be sent to the server in the form of natural language or programming language; the server can perform matching operations on query request information in the form of natural language or programming language.

[0072] For example, keywords can be obtained from the query request information, information categories associated with the keywords can be retrieved from the intelligent knowledge base, and the corresponding information can be extracted from the information category after the corresponding information category is retrieved.

[0073] Optionally, after retrieving the corresponding information category, further retrieval can be performed based on that information category to retrieve more detailed information categories.

[0074] Optionally, after retrieving the corresponding information category, if the information category corresponds to multiple information items, the information with the highest matching degree can be extracted based on the matching degree between the multiple information items and the query request information, or the multiple information items can be extracted in order of matching degree from high to low.

[0075] Step S206: Return the current target information to the terminal corresponding to the query request information, and determine the matching degree information between the query request information and the current target information based on the user feedback data on the target information received by the terminal.

[0076] User feedback data refers to information provided by users to providers of products or services; it can be represented by user opinions, suggestions, evaluations, ratings, etc. Matching information is a measure used to describe the degree of similarity between two objects or data; it can be used to describe the degree of match between a user's query request and the received response.

[0077] For example, the response information matched with the query request information is returned to the terminal, where the user evaluates the degree of matching between the response information and the query request information, and the corresponding matching degree information is extracted based on the user's evaluation result.

[0078] Optionally, users can choose to rate "the two match" or "the two do not match" based on the degree of matching between the response information and the query request information; furthermore, the degree of matching can be quantified, and the degree of matching can be rated based on a value between 0 and 100%.

[0079] Optionally, if a user receives a single response message, each component of that response message can be evaluated separately; if a user receives multiple response messages, each of those response messages can be evaluated separately.

[0080] Step S208: Obtain the semantic training model, determine the training direction of the semantic training model based on the target information with abnormal matching degree information, train the matching degree information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtain the training result, and optimize the information category of each piece of information according to the training result.

[0081] Semantic training models refer to machine learning models trained on large-scale text data to understand the syntax and semantics of a language. Anomalies in matching information refer to a discrepancy between the expected degree of matching between the query request information and the corresponding information; anomalies can be expressed as the number of times poorly relevant information is matched exceeding a preset threshold, or the quantity of poorly relevant information matched exceeding a preset threshold.

[0082] For example, based on the information of abnormal matching degree information, the weights and parameters of the semantic training model are set to determine the training direction of the semantic training model. The matching degree information corresponding to each piece of information in the intelligent knowledge base is trained according to the training direction to obtain the training result. The information category of each piece of information is optimized according to the training result.

[0083] For example, in information with abnormal matching degree information, it can be reflected that the actual correlation between the information and the information category to which it belongs is small, which causes the query request information to match non-corresponding information through the information category. Therefore, the matching degree information can also be represented as the degree of matching between information and the corresponding information category. The training process of the semantic training model can be represented as the process of redetermining the information category of information.

[0084] For example, the training direction of a semantic training model can be represented as a correction direction for information categories based on user preferences. For instance, information with abnormal matching scores may include multiple pieces of information in the same information category matched according to the same query request. Some of these pieces of information have a high degree of matching with the query request, while others have a low degree of matching. The training direction can then be represented as determining a new information category for the information with a high degree of matching and its associated information, and determining a new information category for the information with a low degree of matching and its associated information. The associated information can be represented as information with a large overlap in information categories.

[0085] For example, during the training process, several simulated query request messages are set up. These simulated query request messages are then matched with corresponding information according to a new information category. If the generated matching degree information reaches a preset threshold, the information is determined to correspond to the new information category; if the generated matching degree information does not reach the preset threshold, the information category corresponding to the information is re-determined. The simulated query request messages can be collected using big data technology from query request messages and user feedback data from user groups, or they can be generated using semantic understanding technology.

[0086] Optionally, information categories can be redefined or added during training, and deeper information categories can be further defined on the same topic.

[0087] In the aforementioned information processing method based on semantic understanding, information is classified using semantic understanding technology and hierarchically managed according to information categories. This allows for the rapid retrieval of information that matches the query request based on information categories. Furthermore, the classification of information can be optimized based on user feedback to further improve data retrieval efficiency and achieve the effect of diverting user services.

[0088] In one exemplary embodiment, such as Figure 3As shown, the information category of each piece of information is determined based on the structure of each piece of information in the source database, the correlation between information, and the application scenario, including steps S302 to S306. Wherein:

[0089] Step S302: Based on the structure of the information, confirm the information category as either question-and-answer knowledge or document knowledge.

[0090] Question-and-answer knowledge refers to a knowledge structure organized in the form of answers or suggestions. Document knowledge refers to a knowledge structure organized in the form of files or documents; document knowledge can include information such as various documents, manuals, rules and regulations, and it has the characteristics of universality, comprehensiveness, and long-term effectiveness.

[0091] For example, question-and-answer knowledge focuses more on the correspondence between questions and answers, making it easier to answer users' questions quickly and accurately; document knowledge is more suitable for storing and displaying large amounts of detailed information, from which users can gain in-depth understanding of specific topics.

[0092] Step S304: Based on the dialogue context of the information, identify the information applied to the same conversation as the same information category and as the conversation category.

[0093] Dialogue context refers to a specific environment, scene, or situation in communication; its components may include participants, topic, purpose of dialogue, and context. The same conversation refers to a continuous dialogue or exchange conducted by the same participants or entities over a period of time.

[0094] Participants refer to people or entities in the dialogue, such as users, customers, or robots; topics refer to the themes, issues, or concerns discussed in the dialogue; the purpose of the dialogue refers to the communication goals of the participants, such as obtaining information, solving problems, or providing advice; and context refers to the meaning of specific statements in the dialogue, the preceding and following dialogue content, and background information that influences the dialogue.

[0095] For example, within the same session, the information exchanged between participants has a certain coherence and contextual relationship; therefore, information within the same session category is related.

[0096] Step S306: Information applied to the same scenario is identified as the same information category and is a scenario category.

[0097] An application scenario refers to a specific scene, background, or environment in which a product, service, application, or solution is used or applied. The components of an application scenario may include user behavior, theme, geographical location, user group, and related events.

[0098] User behavior refers to a user's behavioral patterns and interaction history; topic refers to the field, industry, etc., where the information is located; geolocation refers to information such as location and region; user group refers to the audience or target group of the information.

[0099] For example, in the same scenario, information has the same or similar semantic background and tendency, so information of the same scenario category is related.

[0100] Optionally, corresponding scenario-based skill services can be customized according to different scenario needs and integrated into the entire system. After a user matches the corresponding information, they can use the skill services associated with that information.

[0101] In this embodiment, information is classified according to its structure, dialogue context, and application scenario, which enables better organization and management of information and better identification of user intent based on semantic understanding.

[0102] In one exemplary embodiment, such as Figure 4 As shown, based on the query request information, the system retrieves the current target information category that matches the query request information from the intelligent knowledge base, including steps S402 to S406. Wherein:

[0103] Step S402: Extract keywords based on the query request information, and quickly retrieve the first target information category that matches the keywords from the intelligent knowledge base.

[0104] Step S404: Based on the application scenario or topic area involved in the query request information, comprehensively search the intelligent knowledge base for the second target information category that matches the query request information.

[0105] Step S406: The first target information and the second target information constitute the current target information category.

[0106] For example, based on the query request information, the system can determine the user's requirements regarding the quantity and depth of information to be returned, thereby determining the quantity and level of the retrieval. For instance, if the query request information indicates that the user needs a quick response, the system can quickly retrieve the first target information category from the intelligent knowledge base and match the corresponding concise and accurate response information. If the query request information indicates that the user needs a comprehensive response, the system can retrieve the second target information category from the intelligent knowledge base in a multi-level and comprehensive manner, and retrieve the corresponding comprehensive and specific response information.

[0107] For example, the first target information category and the second target information category can be retrieved simultaneously, and the corresponding response information can be output respectively. The response information corresponding to the first target information category can be used as a brief summary text, and the response information corresponding to the second target information category can be used as a detailed description text.

[0108] Optionally, when comprehensively searching the second target information category, knowledge retrieval and matching can be performed for specific scenarios or thematic domains to deeply mine information in the intelligent knowledge base and provide more accurate and comprehensive knowledge services.

[0109] In this embodiment, by performing fast and comprehensive searches on the intelligent knowledge base, the different requirements of users for the number, level, and dimensions of searches during the search process can be met.

[0110] In one exemplary embodiment, such as Figure 5 As shown, based on the user feedback data received by the terminal regarding the target information, the matching degree information between the query request information and the current target information is determined, including steps S502 to S506. Wherein:

[0111] Step S502: If the user reports that the query request information is successfully matched, the matching success rate is determined to be 1, and the matching success rate is used as the matching degree information.

[0112] For example, if the user determines that the returned current response information matches the query request information, then the current matching success rate is determined to be 1.

[0113] Step S504: If the user reports that the query request information fails to match the current target information, the query request information will be retrieved and matched again until the user reports that the query request information matches the current target information.

[0114] For example, if the user determines that the returned current response information does not match the query request information, the user can resend the query request information to re-search and match, or the server can re-search and match the original query request information.

[0115] Step S506: Record the number of times the query request information is matched and the number of information matched each time. Take the sum of the number of information matched each time as the total number of matched information. Take the sum of the number of successfully matched information each time as the total number of successfully matched information. Take the ratio of the total number of successfully matched information to the total number of matched information as the matching success rate. Take the matching success rate as the matching degree information.

[0116] For example, for a query request, multiple response information can be returned simultaneously. The degree of matching between each of the multiple response information and the query request information is determined. The number of response information that is determined to be a match is the number of successfully matched information. The total number of returned response information is the total number of matched information. The ratio of the total number of successfully matched information to the total number of matched information is used as the matching success rate.

[0117] For example, for the same query request information, different response information can be returned multiple times. The degree of matching between each response information and the query request information is judged in turn. The number of response information that is judged to be a match is the number of successfully matched information. The total number of returned response information is the total number of matched information. The ratio of the total number of successfully matched information to the total number of matched information is used as the matching success rate.

[0118] In this embodiment, the degree of matching between query request information and response information is judged by information such as user feedback and user behavior, so that the judgment result is more in line with the user's usage tendency.

[0119] In one exemplary embodiment, such as Figure 6 As shown, the method further includes steps S602 to S604. Wherein:

[0120] Step S602: In the consulting business scenario, based on the query request information, retrieve entity information that matches the query request information from the intelligent knowledge base, and return the entity information to the terminal corresponding to the query request information.

[0121] Among them, entity information refers to information about specific things, places, people, dates, concepts, etc.

[0122] For example, in a consulting business scenario, relevant entity information can be quickly found and answers can be output based on the keywords or questions entered by the user.

[0123] Step S604: In the business scenario of handling inquiries, verify and check the entity information in the query request information, obtain the result information, and return the result information to the terminal corresponding to the query request information.

[0124] For example, in the context of investigation and handling of business, entity information can be verified and checked, and information such as suspected problems or risk points that need attention can be output to help users make effective decisions.

[0125] In this embodiment, by covering scenarios involving consultation and investigation, entity information is processed in a targeted and efficient manner to meet the different business needs of users.

[0126] In one exemplary embodiment, such as Figure 7 As shown, the method further includes step S702. Wherein:

[0127] Step S702: Obtain and analyze user search history, click behavior and feedback information, generate user suggestions, and return the user suggestions to the terminal corresponding to the query request information.

[0128] For example, by analyzing user search behavior and feedback information, user suggestions can be automatically generated to provide users with decision-making directions or even specific implementation plans.

[0129] Optionally, by collecting and analyzing users' search history, click behavior, and feedback information, the accuracy and speed of search results can be improved based on user data.

[0130] In this embodiment, by analyzing user behavior and feedback information to generate user suggestions, the efficiency of user decision-making is improved, as is the efficiency of information interaction between the server and the terminal.

[0131] In one exemplary embodiment, such as Figure 8 As shown, the matching degree information corresponding to each piece of information in the intelligent knowledge base is trained according to the training direction, and steps S802 to S806 are also included.

[0132] Step S802: During the training process, if the matching degree information corresponding to the current information reaches the preset threshold, it is determined that the training of the current information is complete.

[0133] For example, during the training process, it is determined whether the matching success rate of each piece of information has reached a threshold. If the threshold is reached, it is determined that the training of the current information is complete.

[0134] Optionally, each piece of information may correspond to a different threshold, or any piece of information may correspond to the same threshold.

[0135] Step S804: When the number of trained information reaches the preset number, the training process of the semantic training model ends.

[0136] Optionally, training can be performed on a preset amount of information, or on all the information in the intelligent knowledge base.

[0137] Step S806: The semantic training model outputs the trained information categories and updates the information categories corresponding to the target information with abnormal matching degree information to the trained information categories.

[0138] For example, for information that is judged to have a low matching success rate during the response process, the original information category of the information can be updated to the trained information category after training.

[0139] Optionally, other information in the original information category corresponding to this information can be adaptively adjusted, that is, the information category of other information can be adaptively updated after training.

[0140] In this embodiment, by setting a training threshold and a training quantity, the training status of the information is determined, thereby reducing the model training time and improving training efficiency.

[0141] In one embodiment, such asFigure 9 As shown, the above-described information processing method based on semantic understanding may include the following steps:

[0142] On a service management platform that provides knowledge services to users, the platform performs preliminary screening of information in the source database. It determines whether the information is useful based on the reliability of the information source, the accuracy of the information, the timeliness of the information, the relevance of the information, the value of the information, and the retrieval rate. The useful information after preliminary screening is classified at the levels of information structure, dialogue context, and application scenario based on semantic understanding technology. The corresponding information is then associated and stored in an intelligent knowledge base according to the classified information categories. This intelligent knowledge base can serve as a unified knowledge hierarchy management system.

[0143] When a user needs to obtain the electricity cost statistics for the current month, the user sends the query request information to the entity management module of the service management platform through the terminal. The intelligent customer service AI deployed on the service management platform retrieves the information category that matches the query request information from the intelligent knowledge base, extracts the corresponding information from the information category, and returns the information as the response information to the user through the entity management module.

[0144] The entity management module can be used to enable users to operate and manage entity information, thereby realizing the information interaction function between the service management platform and users.

[0145] When a user reports that the response does not match the query request, the feedback is sent to the intelligent customer service AI via the entity management module. The AI ​​then retrieves the relevant information from the intelligent knowledge base and returns the new information as the response to the user. The retrieval process continues until the user reports that the current response matches the query request.

[0146] The intelligent customer service AI records data such as keywords, number of matches, and match success rate for each query. This information is then input into a semantic training model. The semantic training model can analyze problems and propose optimization directions by analyzing data with abnormal matching degrees and issues that occur during the interaction process. It determines the model's architecture and parameters, and uses online learning algorithms to train the information in the intelligent knowledge base, thereby updating the information categories corresponding to each piece of information and adjusting the direction of information acquisition in the intelligent knowledge base.

[0147] Simultaneously, based on the optimized intelligent knowledge base, the intelligent customer service AI is analyzed and upgraded, followed by testing and debugging to identify and promptly fix relevant issues. This improves the accuracy and speed of the AI's response matching, thereby ensuring the quality of the response service and optimizing service effectiveness. During the optimization of the intelligent knowledge base, the AI's adaptive, dynamic, and incremental self-learning capabilities are achieved, ensuring the system's gradual optimization.

[0148] In addition, this intelligent customer service AI can provide services for intelligent IVR, online customer service AI, etc. through open external interfaces and debugging, and provide intelligent empowerment for different channels to achieve intelligent response, multi-turn conversation, and enrich the application scenarios of intelligent knowledge base.

[0149] Specifically, when a user asks the same question four times, among the four responses A, B, C, and D received in sequence, only the last response D is reported by the user as matching the query request information. In this case, the matching success rate in the interaction process is 25%.

[0150] Based on the data showing a low matching success rate, the training direction of the semantic training model is determined as follows: redefine the information categories of information A, B, C, and D, so that when users submit a query request for "get this month's electricity cost statistics" or other similar information, they can more accurately retrieve the corresponding information category, thus favoring the extraction of information D and similar information, and avoiding the extraction of information A, B, C, and similar information.

[0151] During this training process, a new information category is determined for information D. A query request is simulated. If the query request matches information D by retrieving the new information category, information D is associated with the new information category and stored in the intelligent knowledge base. Simultaneously, other information with a high degree of overlap with the original information category of information D is trained, and the information categories of these other information are adaptively adjusted.

[0152] When updating information categories in the intelligent knowledge base, the information can be further filtered. Therefore, the process of updating information categories is also a process of continuous filtering and updating of information.

[0153] In addition, when users need to obtain the electricity cost statistics for this month, the intelligent customer service AI can not only provide the electricity cost data for this month, but also grant the user access to view the electricity cost data for previous months, annual electricity cost data, and monthly electricity consumption fluctuation comparison table, thereby realizing the function of intelligently judging the user's intention and realizing multi-round interaction with the user.

[0154] After the user has finished viewing the electricity bill data, the intelligent customer service AI compares the user's monthly electricity bill with the user's average monthly electricity bill and reminds the user if there is any abnormality in electricity usage. If there is an abnormality, the AI ​​will give the user suggestions based on the user's average peak electricity usage time for the month, such as whether high-energy-consuming appliances were used during that time.

[0155] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0156] Based on the same inventive concept, this application also provides a semantic understanding-based information processing apparatus for implementing the semantic understanding-based information processing method described above. The solution provided by this apparatus is similar to the implementation scheme described in the above method; therefore, the specific limitations in one or more embodiments of the semantic understanding-based information processing apparatus provided below can be found in the limitations of the semantic understanding-based information processing method described above, and will not be repeated here.

[0157] In one exemplary embodiment, such as Figure 10 As shown, a semantic understanding-based information processing device is provided, comprising: a management module 1002, a matching module 1004, an entity module 1006, and a training module 1008, wherein:

[0158] The management module 1002 is used to determine the information category of each piece of information based on semantic understanding technology, according to the structure, dialogue context and application scenario of each piece of information in the source database, and to associate and store the corresponding target information in the intelligent knowledge base according to the target information category.

[0159] Matching module 1004 is used to retrieve the current target information category that matches the query request information from the intelligent knowledge base based on the query request information, and extract the corresponding current target information based on the current target information category;

[0160] Entity module 1006 is used to return the current target information to the terminal corresponding to the query request information, and determine the matching degree information between the query request information and the current target information based on the user feedback data on the target information received by the terminal.

[0161] The training module 1008 is used to acquire the semantic training model, determine the training direction of the semantic training model based on the target information of the matching degree information anomaly, train the matching information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtain the training results, and optimize the information category of each piece of information according to the training results.

[0162] In an exemplary embodiment, the management module 1002 further includes: a structure category unit, a dialogue context category unit, and an application scenario category unit, wherein:

[0163] The structural category unit is used to identify the information category as either question-and-answer knowledge or document knowledge based on the information's structure.

[0164] The dialogue context category unit is used to identify information applied to the same conversation as the same information category and conversation category based on the dialogue context of the information.

[0165] The application scenario category unit identifies information that is applied to the same scenario as the same information category, and is also a scenario category.

[0166] In an exemplary embodiment, the matching module 1004 further includes: a first retrieval unit, a second retrieval unit, and a set unit, wherein:

[0167] The first retrieval unit is used to extract keywords based on the query request information and quickly retrieve the first target information category that matches the keywords from the intelligent knowledge base.

[0168] The second retrieval unit is used to comprehensively retrieve the second target information category that matches the query request information from the intelligent knowledge base, based on the application scenario or topic area involved in the query request information.

[0169] The set unit is used to combine the first target information and the second target information into the current target information category.

[0170] In an exemplary embodiment, entity module 1006 further includes: a first determination unit and a second determination unit, wherein:

[0171] The first determination unit is used to determine the matching success rate as 1 if the user reports that the query request information is successfully matched, and to use the matching success rate as the matching degree information.

[0172] The second determination unit is used to re-search and match the query request information if the user reports that the query request information fails to match the current target information, until the user reports that the query request information matches the current target information; it is also used to record the number of times the query request information is matched, the number of information matched each time, the sum of the number of information matched each time as the total number of matched information, the sum of the number of successfully matched information each time as the total number of successfully matched information, the ratio of the total number of successfully matched information to the total number of matched information as the matching success rate, and the matching success rate as the matching degree information.

[0173] In one exemplary embodiment, the device further includes: a consultation service module and an investigation service module, wherein:

[0174] The consultation service module is used in consultation service scenarios to retrieve entity information that matches the query request information from the intelligent knowledge base and return the entity information to the terminal corresponding to the query request information.

[0175] The investigation and handling module is used to verify and check the entity information in the query request information in the investigation and handling business scenario, obtain the result information, and return the result information to the terminal corresponding to the query request information.

[0176] In an exemplary embodiment, the device further includes: a first training determination module, a second training determination module, and a training result output module, wherein:

[0177] The user data analysis module is used to acquire and analyze user search history, click behavior and feedback information, generate user suggestions, and return the user suggestions to the terminal corresponding to the query request information.

[0178] In an exemplary embodiment, the training module 1008 further includes: a first determination unit, a second determination unit, and a training result generation unit, wherein:

[0179] The first training determination unit is used to determine that the training of the current information is complete when the matching degree information corresponding to the current information reaches a preset threshold during the training process.

[0180] The second training decision unit is used to end the training process of the semantic training model when the amount of information that has been trained reaches a preset amount.

[0181] The training result generation unit is used to output the trained information category of the semantic training model and update the information category of the target information with abnormal matching degree information to the trained information category.

[0182] The modules in the aforementioned semantic understanding-based information processing device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can invoke and execute the operations corresponding to each module.

[0183] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11 As shown, this computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores data such as information categories, response information, and query request information. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements an information processing method based on semantic understanding.

[0184] Those skilled in the art will understand that Figure 11 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0185] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0186] Based on semantic understanding technology, the information category of each piece of information is determined according to the structure, dialogue context and application scenario of each piece of information in the source database, and the corresponding target information is associated and stored in the intelligent knowledge base according to the target information category.

[0187] Based on the query request information, retrieve the current target information category that matches the query request information from the intelligent knowledge base, and extract the corresponding current target information based on the current target information category;

[0188] The current target information is returned to the terminal corresponding to the query request information. Based on the user feedback data on the target information received by the terminal, the matching degree information between the query request information and the current target information is determined.

[0189] Obtain the semantic training model, determine the training direction of the semantic training model based on the target information with abnormal matching degree information, train the matching degree information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtain the training results, and optimize the information category of each piece of information according to the training results.

[0190] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0191] Based on the structure of the information, identify whether the information is classified as question-and-answer knowledge or document knowledge.

[0192] Based on the dialogue context of the information, information applied to the same conversation is identified as the same information category and is also a conversation category;

[0193] Based on the application scenario of the information, information applied to the same scenario is identified as the same information category and is also a scenario category.

[0194] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0195] Extract keywords from the query request information and quickly retrieve the first target information category that matches the keywords from the intelligent knowledge base;

[0196] Based on the application scenario or topic area involved in the query request information, comprehensively search the intelligent knowledge base for the second target information category that matches the query request information;

[0197] The first target information and the second target information constitute the current target information category.

[0198] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0199] If the user reports that the query request information is successfully matched, the matching success rate is determined to be 1, and the matching success rate is used as the matching degree information.

[0200] If the user reports that the query request information fails to match the current target information, the query request information will be retrieved and matched again until the user reports that the query request information matches the current target information.

[0201] Record the number of times the query request information is matched, the number of information matched each time, the sum of the number of information matched each time as the total number of matched information, the sum of the number of successfully matched information each time as the total number of successfully matched information, the ratio of the total number of successfully matched information to the total number of matched information as the matching success rate, and the matching success rate as the matching degree information.

[0202] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0203] In the context of consulting services, based on the query request information, the system retrieves entity information that matches the query request information from the intelligent knowledge base and returns the entity information to the terminal corresponding to the query request information.

[0204] In the context of business investigation, the entity information in the query request is verified and checked to obtain the result information, which is then returned to the terminal corresponding to the query request.

[0205] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0206] It acquires and analyzes user search history, click behavior, and feedback information, generates user suggestions, and returns the user suggestions to the terminal corresponding to the query request information.

[0207] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0208] During training, if the matching degree information corresponding to the current information reaches the preset threshold, the training of the current information is considered complete.

[0209] When the amount of trained information reaches the preset amount, the training process of the semantic training model ends.

[0210] The semantic training model outputs the trained information categories and updates the information categories corresponding to target information with abnormal matching information to the trained information categories.

[0211] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0212] Based on semantic understanding technology, the information category of each piece of information is determined according to the structure, dialogue context and application scenario of each piece of information in the source database, and the corresponding target information is associated and stored in the intelligent knowledge base according to the target information category.

[0213] Based on the query request information, retrieve the current target information category that matches the query request information from the intelligent knowledge base, and extract the corresponding current target information based on the current target information category;

[0214] The current target information is returned to the terminal corresponding to the query request information. Based on the user feedback data on the target information received by the terminal, the matching degree information between the query request information and the current target information is determined.

[0215] Obtain the semantic training model, determine the training direction of the semantic training model based on the target information with abnormal matching degree information, train the matching degree information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtain the training results, and optimize the information category of each piece of information according to the training results.

[0216] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0217] Based on the structure of the information, identify whether the information is classified as question-and-answer knowledge or document knowledge.

[0218] Based on the dialogue context of the information, information applied to the same conversation is identified as the same information category and is also a conversation category;

[0219] Based on the application scenario of the information, information applied to the same scenario is identified as the same information category and is also a scenario category.

[0220] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0221] Extract keywords from the query request information and quickly retrieve the first target information category that matches the keywords from the intelligent knowledge base;

[0222] Based on the application scenario or topic area involved in the query request information, comprehensively search the intelligent knowledge base for the second target information category that matches the query request information;

[0223] The first target information and the second target information constitute the current target information category.

[0224] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0225] If the user reports that the query request information is successfully matched, the matching success rate is determined to be 1, and the matching success rate is used as the matching degree information.

[0226] If the user reports that the query request information fails to match the current target information, the query request information will be retrieved and matched again until the user reports that the query request information matches the current target information.

[0227] Record the number of times the query request information is matched, the number of information matched each time, the sum of the number of information matched each time as the total number of matched information, the sum of the number of successfully matched information each time as the total number of successfully matched information, the ratio of the total number of successfully matched information to the total number of matched information as the matching success rate, and the matching success rate as the matching degree information.

[0228] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0229] In the context of consulting services, based on the query request information, the system retrieves entity information that matches the query request information from the intelligent knowledge base and returns the entity information to the terminal corresponding to the query request information.

[0230] In the context of business investigation, the entity information in the query request is verified and checked to obtain the result information, which is then returned to the terminal corresponding to the query request.

[0231] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0232] It acquires and analyzes user search history, click behavior, and feedback information, generates user suggestions, and returns the user suggestions to the terminal corresponding to the query request information.

[0233] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0234] During training, if the matching degree information corresponding to the current information reaches the preset threshold, the training of the current information is considered complete.

[0235] When the amount of trained information reaches the preset amount, the training process of the semantic training model ends.

[0236] The semantic training model outputs the trained information categories and updates the information categories corresponding to target information with abnormal matching information to the trained information categories.

[0237] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0238] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0239] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0240] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An information processing method based on semantic understanding, characterized in that, include: Based on semantic understanding technology, the information category of each piece of information is determined according to the structure, dialogue context and application scenario of each piece of information in the source database, and the corresponding target information is associated and stored in the intelligent knowledge base according to the target information category. Based on the query request information, the system retrieves the current target information category that matches the query request information from the intelligent knowledge base, and extracts the corresponding current target information based on the current target information category. The current target information is returned to the terminal corresponding to the query request information, and the matching degree information between the query request information and the current target information is determined based on the user feedback data on the target information received by the terminal. A semantic training model is obtained. The training direction of the semantic training model is determined based on target information with abnormal matching degree information. The matching degree information corresponding to each piece of information in the intelligent knowledge base is trained according to the training direction to obtain training results. The information category of each piece of information is optimized based on the training results. The training direction of the semantic training model includes a correction direction for the information category based on user preferences. Information with abnormal matching degree information includes multiple pieces of information in the same information category matched according to the same query request. The training direction includes determining information with a matching degree higher than a preset value and its associated information as a new information category, and determining information with a matching degree lower than a preset value and its associated information as a new information category. The associated information includes information whose information category overlap meets preset conditions. During the training process of the semantic training model, several simulated query request information are set. The simulated query request information is matched with corresponding information through new information categories. If the generated matching degree information does not reach a preset threshold, the information category corresponding to that information is redefined. Simulated query request information is collected from query request information and user feedback data from user groups through big data technology or generated through semantic understanding technology.

2. The method according to claim 1, characterized in that, The process of determining the information category of each piece of information based on the structure of each piece of information in the source database, the correlation between information, and the application scenario includes: Based on the structure of the information, identify whether the information is classified as question-and-answer knowledge or document knowledge. Based on the dialogue context of the information, information applied to the same conversation is identified as the same information category and is also a conversation category; Based on the application scenario of the information, information applied to the same scenario is identified as the same information category and is also a scenario category.

3. The method according to claim 1, characterized in that, The step of retrieving the current target information category that matches the query request information from the intelligent knowledge base includes: Based on the query request information, keywords are extracted, and the first target information category matching the keywords is quickly retrieved from the intelligent knowledge base. Based on the application scenario or topic area involved in the query request information, a second target information category that matches the query request information is comprehensively retrieved in the intelligent knowledge base; The first target information and the second target information constitute the current target information category.

4. The method according to claim 1, characterized in that, The step of determining the matching degree information between the query request information and the current target information based on user feedback data on the target information received by the terminal includes: If the user reports that the query request information is successfully matched, the matching success rate is determined to be 1, and the matching success rate is used as the matching degree information. If the user reports that the query request information fails to match the current target information, the query request information will be retrieved and matched again until the user reports that the query request information matches the current target information. Record the number of times the query request information is matched and the number of information matched each time. The sum of the number of information matched each time is taken as the total number of matched information. The sum of the number of successfully matched information each time is taken as the total number of successfully matched information. The ratio of the total number of successfully matched information to the total number of matched information is taken as the matching success rate. The matching success rate is taken as the matching degree information.

5. The method according to claim 1, characterized in that, The method further includes: In the context of consulting services, based on the query request information, entity information matching the query request information is retrieved from the intelligent knowledge base, and the entity information is returned to the terminal corresponding to the query request information. In the context of investigation and handling of business, the entity information in the query request information is verified and checked to obtain result information, and the result information is returned to the terminal corresponding to the query request information.

6. The method according to claim 1, characterized in that, The method further includes: The system acquires and analyzes user search history, click behavior, and feedback information to generate user suggestions, which are then returned to the terminal corresponding to the query request information.

7. The method according to claim 1, characterized in that, The step of training the matching degree information corresponding to each piece of information in the intelligent knowledge base according to the training direction includes: During the training process, if the matching degree information corresponding to the current information reaches a preset threshold, it is determined that the training of the current information is complete. When the number of trained information reaches a preset number, the training process of the semantic training model ends. The semantic training model outputs the trained information category and updates the information category corresponding to the target information with abnormal matching degree information to the trained information category.

8. An information processing device based on semantic understanding, characterized in that, The device includes: The management module is used to determine the information category of each piece of information based on semantic understanding technology, according to the structure of each piece of information in the source database, the dialogue context and the application scenario, and to associate and store the corresponding target information in the intelligent knowledge base according to the target information category. The matching module is used to retrieve the current target information category that matches the query request information from the intelligent knowledge base based on the query request information, and extract the corresponding current target information based on the current target information category; The entity module is used to return the current target information to the terminal corresponding to the query request information, and determine the matching degree information between the query request information and the current target information based on the user feedback data on the target information received by the terminal. The training module is used to acquire a semantic training model, determine the training direction of the semantic training model based on target information with abnormal matching degree information, train the matching information corresponding to each piece of information in the intelligent knowledge base according to the training direction, obtain training results, and optimize the information category of each piece of information according to the training results. The training direction of the semantic training model includes a correction direction for the information category based on user preferences. Information with abnormal matching degree information includes multiple pieces of information in the same information category matched according to the same query request. The training direction includes determining information with a matching degree higher than a preset value and its associated information as a new information category, and determining information with a matching degree lower than a preset value and its associated information as a new information category. The associated information includes information whose information category overlap range meets preset conditions. During the training process of the semantic training model, several simulated query request information are set, and the simulated query request information is matched with corresponding information through new information categories. If the generated matching degree information does not reach a preset threshold, the information category corresponding to the information is re-determined. The simulated query request information is collected from query request information and user feedback data from user groups using big data technology or generated through semantic understanding technology.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images