Multi-dimensional adaptive corpus demand matching method and system

Abstract

The embodiment of the invention relates to the field of data processing, and discloses a multi-dimensional adaptive corpus demand matching method. The method comprises the following steps: if user demand information is received, determining demand feature vectors and dimensions of the demand feature vectors; mapping in a preset tag library according to the dimension of each demand feature vector, and determining a tag type corresponding to the dimension of each demand feature vector; according to a preset demand matching formula, calculating a matching degree between the demand feature vector and each tag under each tag type in a preset tag library; and if the matching degree exceeding the preset matching degree threshold exists, taking the label corresponding to the matching degree exceeding the preset matching degree threshold as a target label, generating response information according to the target label, and feeding back the response information to the user. The technical problems that a large amount of manpower and time need to be consumed when user requirements are judged manually, the efficiency is low when a large amount of data is processed, and complex and changeable user requirements are difficult to efficiently and accurately process can be solved at least.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a multi-dimensional adaptive corpus demand matching method and system. Background Art

[0002] With the rapid development of information technology and the increasing diversification of user needs, how to accurately understand and meet the personalized needs of users has become an important challenge faced by enterprises and service providers. Traditional corpus demand matching methods often rely on manual analysis or simple rule matching, that is, staff process the collected user demand corpus based on their own experience and established rules to generate corresponding matching results.

[0003] However, the inventors have found that there are at least the following technical problems in the related technology: manual analysis requires a lot of manpower and time, and is inefficient, especially when processing a large amount of complex demand corpus, and it is difficult to efficiently and accurately adapt to the ever-changing user needs. Summary of the invention

[0004] One purpose of the present application is to provide a multi-dimensional adaptive corpus demand matching method, at least to solve the problem that manual judgment of user corpus needs consumes a lot of manpower and time, especially when processing large amounts of data, the efficiency is low, and it is difficult to efficiently and accurately handle complex and changeable user needs.

[0005] To achieve the above objectives, some embodiments of the present application provide the following aspects:

[0006] In a first aspect, some embodiments of the present application further provide a multi-dimensional adaptive corpus demand matching method, the method comprising:

[0007] If user demand information is received, historical user interaction data and current context information are obtained, and demand feature vectors and dimensions of each demand feature vector are determined according to the user demand information, historical user interaction data and current context information;

[0008] Mapping is performed in a preset label library according to the dimension of each demand feature vector to determine the label type corresponding to the dimension of each demand feature vector;

[0009] According to the preset demand matching formula, the matching degree between the demand feature vector and each label under each label type in the preset label library is calculated; wherein the preset demand matching formula is:

[0010]

[0011] Among them, M fis the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product calculates the similarity between two vectors. The larger the dot product,

[0012] It indicates that the higher the matching degree between the demand feature vector and the label;

[0013] If there is a match degree exceeding the preset match degree threshold, the tag corresponding to the match degree exceeding the preset match degree threshold is used as the target tag, response information is generated according to the target tag, and the response information is fed back to the user.

[0014] According to a second aspect of the present application, a multi-dimensional adaptive corpus demand matching system is provided, the system comprising:

[0015] A demand feature vector determination module is used to, if receiving user demand information, obtain user historical interaction data and current context information, and determine demand feature vectors and dimensions of each demand feature vector according to the user demand information, user historical interaction data and current context information;

[0016] A label type determination module is used to map the dimension of each demand feature vector in a preset label library to determine the label type corresponding to the dimension of each demand feature vector;

[0017] The matching degree calculation module is used to calculate the matching degree between the demand feature vector and each label under each label type in the preset label library according to the preset demand matching formula; wherein the preset demand matching formula is:

[0018]

[0019] Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product calculates the similarity between two vectors. The larger the dot product,

[0020] It indicates that the higher the matching degree between the demand feature vector and the label;

[0021] The feedback module is used to, if there is a matching degree exceeding a preset matching degree threshold, use the tag corresponding to the matching degree exceeding the preset matching degree threshold as the target tag, generate response information according to the target tag, and feed back the response information to the user.

[0022] Compared with the related art, in the scheme provided by the embodiment of the present application, if user demand information is received, user historical interaction data and current scenario information are obtained, and the demand feature vector and the dimension of each demand feature vector are determined according to the user demand information, user historical interaction data and current scenario information; the dimension of each demand feature vector is mapped in the preset tag library to determine the tag type corresponding to the dimension of each demand feature vector; according to the preset demand matching formula, the matching degree of each tag under each tag type in the preset tag library is calculated; if there is a matching degree exceeding the preset matching degree threshold, the tag corresponding to the matching degree exceeding the preset matching degree threshold is used as the target tag, and the response information is generated according to the target tag, and the response information is fed back to the user. Through the above-mentioned multi-dimensional adaptive corpus demand matching method, the recommendation obtained by the user can be more accurate, in line with personal habits and current needs, thereby improving satisfaction. As user behavior and environment change, the recommended content can be adjusted to maintain the timeliness and relevance of the recommendation. At the same time, the matching process is efficient, and the most suitable tag type can be quickly found, the search efficiency is improved, and the target tag that best meets the user's needs is automatically screened out, which can improve the accuracy of the recommendation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0024] Figure 1 is an exemplary flow chart of a multi-dimensional adaptive corpus demand matching method provided in Example 1 of the present application;

[0025] Figure 2 It is a schematic diagram of the contents of the preset tag library provided in Example 1 of the present application;

[0026] Figure 3 This is an exemplary structural diagram of a multi-dimensional adaptive corpus demand matching system provided in Example 4 of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0028] First embodiment

[0029] The first embodiment of the present application relates to a multi-dimensional adaptive corpus demand matching method. Figure 1 As shown, the method may include the following steps:

[0030] S101: If user demand information is received, user historical interaction data and current context information are obtained, and demand feature vectors and dimensions of each demand feature vector are determined according to the user demand information, user historical interaction data and current context information.

[0031] First, the use scenario of this solution can be a scenario in which corresponding tags are matched in a tag library according to user demand information, and response information is generated and sent to the user.

[0032] Based on the above usage scenarios, it can be understood that the execution subject of this application can be multi-dimensional adaptive corpus demand matching, and no excessive limitations are made here.

[0033] In this solution, user demand information can be a corpus demand or request clearly expressed by the user at a certain point in time. It usually appears in natural language and describes the user's goal or desired result. It is the core data for the system to understand the user's current expectations. By parsing this information, the system can identify the user's intentions, the type of product or service expected, etc. For example, the user enters "Please translate this article into French" and attaches the text to be translated. The system needs to provide accurate translation services based on this demand.

[0034] User historical interaction data can be a record of the user's past interactions with the system related to corpora. These data reflect the user's preferences, habits, and needs for specific corpora. These data reflect the user's preferences and habits. User historical interaction data can include search history: specific corpus types or topics that the user has searched for. For example, the user frequently searched for literature or articles related to "technical development of natural language processing" in the past. Click records: corpus resources that the user has browsed or clicked in the system. For example, the user often clicked on research papers or tutorials related to "machine learning" in the past. Corpus usage records: specific corpus content used by the user, such as the frequency of citing certain literature, or templates or sentence structures that the user has repeatedly used. Favorited or marked content: corpora that the user has collected or marked as important, such as specific research literature, sentence structures, or article paragraphs. For example, the user has searched and read technical articles on "generative adversarial networks" many times before, and collected some classic research literature on generative adversarial networks. These historical records reflect the user's strong interest in and related needs for generative adversarial networks. When the user requests corpora related to generative adversarial networks again, the system will prioritize related or more advanced research materials based on these historical interaction data to meet the user's specific corpus needs.

[0035] Current context information can be specific information about the user's current usage environment. This information can include the user's research field, the topic of the project being processed, the type of device used, the time period, etc. This context information helps the system match and recommend relevant corpora more accurately. For example, in an academic research scenario, if the user is currently writing a paper on artificial intelligence, the system may prioritize the latest research papers and technical reports related to AI. In a translation scenario, if the user is currently processing a technical document translation, the system may recommend relevant professional terminology vocabularies and reference materials.

[0036] The demand feature vector can be a multidimensional vector generated by integrating user demand information, user historical interaction data, and current situation information. This vector represents the specific characteristics of the user in different corpus demand dimensions and is used to match corpus resources or labels in the system. For example, in a language learning scenario, the user's demand information is "listening materials for practicing English", historical data shows that the user mainly studied business English in the past, and the current situation is that the user is preparing for an English interview. The demand feature vector may include dimensions such as "business English", "listening practice", and "interview skills".

[0037] The dimensions of a demand feature vector may refer to specific attributes or features represented by each value in the demand feature vector. These dimensions represent various aspects of user demand information, which are related to user preferences, historical behaviors, or current scenarios. For example, in a language learning scenario, the user's corpus demand is "looking for teaching materials to improve spoken English", and the dimensions may include "language level" (such as intermediate), "teaching material type" (such as oral practice), "content difficulty" (such as advanced conversation), and "application scenario" (such as daily communication or business occasions). These dimensions help the system understand the specific attributes of user needs in order to recommend the most appropriate learning materials.

[0038] It can receive and parse the user's current input, such as search terms or voice commands, and extract the key elements of the user's needs. It can retrieve and analyze the user's historical behavior data, and identify the user's preferences and habits in the user's historical interaction data based on the user's demand information. For example, in the corpus demand matching scenario, after receiving the user's demand information, the system will parse the user's current input (such as search terms, voice commands) and extract the key elements of the user's needs. Subsequently, the system will retrieve and analyze the user's historical behavior data, which reflects the user's preferences and habits. For example, if the user has selected a certain type of literature in a specific field many times in the past, this preference will affect the weight distribution of the relevant dimensions in the demand feature vector. Then, the system will further modify the demand feature vector in combination with the current situational information, such as the user's research topic, current task requirements, etc. Finally, the above information is integrated to generate a multi-dimensional feature vector, which will play a key role in the subsequent corpus matching process. Each feature will be quantified and mapped to a dimension in the feature vector. Specifically, the user demand information can be parsed through NLP technology to extract keywords, emotions, demand intentions, etc., and converted into numerical representations. All quantified feature values ​​are arranged in the preset dimension order to form a demand feature vector. For example, the user's demand information is "find the latest machine learning review articles", and the user's historical interaction data shows that they frequently consulted review articles from top conferences in the past and prefer resources that include code implementation. The current context information is that the user is reviewing research in the field of machine learning. The features of each dimension are quantified as follows:

[0039] Topic dimension: The demand involves "Overview of Machine Learning". Assume that a vector is used to represent the topic features, where "Overview of Machine Learning" can be quantified as 0.9 (0 means irrelevant and 1 means highly relevant).

[0040] Source dimension: Users prefer top conference articles, and use a vector to represent the source feature, where "top conference" can be quantified as 0.8 (0 represents a low-impact source and 1 represents a high-impact source).

[0041] Additional content dimension: Users prefer articles with code implementation. The additional content feature is represented by a vector, and “with code” can be quantified as 0.7 (0 means no code, 1 means with code).

[0042] Time dimension: The demand for the latest article can be quantified as 1.0 (0 represents old articles and 1 represents the latest article).

[0043] Therefore, the demand feature vector can be expressed as [0.9, 0.8, 0.7, 1.0], where each dimension represents the topic, source, additional content, and time. This demand feature vector will be used to match in the preset tag library to find the corpus resource that best meets the user's needs.

[0044] Based on the above technical solution, optionally, before obtaining the user's historical interaction data and current context information, the method further includes:

[0045] Determining whether the user demand information is complete;

[0046] If the user demand information is incomplete, a request for additional information is sent to the user;

[0047] Receive the supplementary demand information sent by the user, and determine whether the supplementary demand information is complete;

[0048] If the demand supplementary information is complete, updating the user demand information according to the demand supplementary information;

[0049] Accordingly, obtaining the user's historical interaction data and current context information, and determining the demand feature vector according to the user's demand information, the user's historical interaction data and the current context information, includes:

[0050] The demand feature vector is determined based on the updated user demand information, user historical interaction data and current context information.

[0051] In this solution, the information supplement request can be a prompt or inquiry issued by the system to the user, with the purpose of collecting necessary information that the user has not yet provided. These requests can be questions for specific issues, multiple-choice questions, or open-ended questions that require users to provide more details. For example, a user enters a requirement in the corpus demand matching system: "I need to find articles that meet my needs." The system can issue an information supplement request to collect more details, such as a question for a specific issue: "Do you have any special requirements for the topic of the article? For example, technology, history, or entertainment, etc.?", or a multiple-choice question: "Please select the type of article you are interested in from the following options: 1) Technology 2) History 3) Entertainment 4) Health" or an open-ended question: "Please provide more information about the topic or specific article type you are interested in."

[0052] The information provided by the user can be analyzed first to check whether it meets all the key dimensions required to perform demand analysis or generate recommendations. If some necessary information is missing, the system marks this information as "incomplete". Generate appropriate questions or prompts based on the missing information, and send the request to the user through the user interface (such as a chat window, pop-up dialog box, etc.). The supplementary request should be clear and specific to help the user clearly understand the information that needs to be supplemented. After receiving the supplementary request, the user enters the required additional information. The system receives this information and performs another completeness check to ensure that the newly supplemented information is sufficient to make up for the previous missing parts. If the system confirms that the supplementary information is complete, it will integrate this new information into the user's initial requirements and update the requirement feature vector.

[0053] In this solution, requesting the user to complete the information when the user demand is incomplete can enable more accurate processing and response, thereby improving the overall user experience and service quality.

[0054] On the basis of the above technical solution, optionally, if user demand information is received, user historical interaction data and current context information are obtained, and a demand feature vector is determined according to the user demand information, user historical interaction data and current context information, including:

[0055] Determine the semantic information, intent information and keyword information of user demand information;

[0056] Determine an initial feature vector based on the semantic information, intention information and keyword information;

[0057] Matching the target historical behavior in the user historical interaction data according to the semantic information, the intention information, and the keyword information, and determining a behavior vector of the target historical behavior;

[0058] A scenario vector is determined according to current scenario information, and a demand feature vector is determined according to the scenario vector, the initial feature vector and the behavior vector.

[0059] In this solution, semantic information can refer to the meaning and context extracted from the user demand information. For example, for the user demand "find the latest research on machine learning", the semantic information includes "machine learning" and "latest research".

[0060] Intent information can be the user's specific needs and expected behaviors. Intent information can involve the goals that the user wants to achieve. For example, if the user wants to obtain the latest machine learning research results, the intent information is "obtain the latest research."

[0061] The keyword information may be keywords extracted from user requirements, which are directly related to the user requirements. For example, for the requirement "find the latest research on machine learning", the keyword information includes "machine learning" and "latest research".

[0062] The initial feature vector can be a vector representation generated based on semantic information, intent information, and keyword information. This vector captures the core features of user needs for subsequent processing. For example, based on the extraction of "machine learning" and "latest research", a feature vector such as [0.9, 0.8, 0.7] can be generated, where each dimension represents a different feature, such as the field (machine learning) and the novelty of the research (latest research).

[0063] The target historical behavior can be the behavior or preference that the user has shown in past interactions, which is used to help determine the user's demand characteristics. For example, if the user has frequently consulted machine learning related literature or research in the past, these behaviors constitute the target historical behavior.

[0064] The behavior vector can be a vector representation extracted from the target historical behavior, indicating the user's historical preferences and behavior patterns.

[0065] The context vector may represent the influencing factors of the current context, which may include time, location, environment, etc. For example, if the current context is that the user is at an academic conference, the context vector may be [1, 0.9], indicating the high influence and relevance of the academic activity.

[0066] This information can be extracted from the user's demand description using natural language processing (NLP) technology. For example, the semantic information extracted from "finding the latest research on machine learning" includes "machine learning" and "latest research". Based on the extracted semantic information, intent information, and keyword information, this information is quantified into a feature vector. For example, the user intent is "to obtain the latest research results in machine learning." "Machine learning" and "latest research" are keyword information. For "latest research in machine learning", the initial feature vector can be [0.9, 0.8, 0.7], where each dimension represents a different feature, such as the research field (machine learning) and the novelty of the research (latest research). Find behaviors related to the current needs in the user's historical interaction data. For example, the user may have often viewed machine learning-related literature or participated in related discussions in the past. These behaviors constitute the target historical behaviors. These target historical behaviors are converted into vector representations, such as [0.85, 0.9], which represent the user's interest in the field of machine learning and attention to the latest research. Then based on the current The previous context generates a context vector. For example, if the current context is that the user is at an academic conference, the context vector can be [1, 0.9], indicating the high impact and relevance of the academic activity. Based on the initial feature vector, the behavior vector, and the context vector, a final demand feature vector is generated through a weighted combination or calculation method. For example, the demand feature vector can be a weighted sum, average, or other combination of these vectors. For example, the initial feature vector [0.9, 0.8, 0.7], the behavior vector [0.85, 0.9], and the context vector [1, 0.9] are weightedly combined to generate a final demand feature vector, such as [0.92, 0.85, 0.78].

[0067] In this scheme, by comprehensively considering semantic information, intent information, keyword information, historical behavior and current context, the generated demand feature vector can more accurately reflect the actual needs of users, help improve the accuracy of the recommendation system, and make the recommendation results more in line with user expectations.

[0068] S102, mapping is performed in a preset label library according to the dimension of each demand feature vector, and the label type corresponding to the dimension of each demand feature vector is determined.

[0069] The preset tag library can be a database containing a large number of corpus tags and their corresponding features, which is used to store and organize tags related to different demand features. Each tag library is defined and designed according to different application scenarios and demand features. The tags in the tag library are classified into various tag types, such as subject, source, additional content, time, etc. The tag library plays the role of a mapping bridge in the system, matching the user's demand feature vector with the actual resources. In this solution, Figure 2 is a schematic diagram of the contents of the preset tag library provided in the embodiment of the present application, such as Figure 2 As shown, the preset tag library can be a hierarchical classification system for organizing and managing content in different fields. Each tag library contains a first-level classification and a second-level classification. The first-level classification represents a broad topic or field, and the second-level classification further refines the specific content category under the first-level classification.

[0070] The tag type may refer to the category or attribute corresponding to each dimension of the demand feature vector in the tag library. The tag library may contain predefined tags, each with different feature values ​​and attributes. The tag type may be a classification of specific attributes or features in the tag library. For example, in an academic document recommendation system, the tag library may contain subject tags: such as "machine learning", "natural language processing", "data science", etc., and each tag type may have a specific tag value (such as "deep learning", "supervised learning"). Source tags: such as "top conferences", "journals", "white papers", etc., this tag type helps determine the quality of the source of the document. Additional content tags: such as "contains code", "has data sets", "has detailed mathematical derivations", etc., this tag type identifies whether the document comes with additional content that the user is interested in. Time tags: such as "the last year", "the past five years", "classic literature", etc., this tag type helps identify the timeliness of the document. These tag types correspond one-to-one to the dimensions in the user demand feature vector, helping the system find the corpus resources that best meet the user's needs in the tag library. For example, if the value of the "topic" dimension in the user demand feature vector is "0.9" (representing a highly relevant machine learning review), the system will prioritize searching for tags that are highly relevant to "machine learning" under the "topic tag" type to match the most suitable literature. In this scheme, the first-level tag type can be the broadest classification level, usually representing a large category or a broad subject area. The first-level tag type is used to organize the main fields or topics and is the top-level classification of the tag library. The second-level tag type can be a sub-classification under the first-level tag type, which is used to further refine and concretize the content of the first-level tag. The second-level tag type provides a more refined classification of a certain field or topic and is the next level classification of the tag library.

[0071] According to the dimensions of the demand feature vector, the tag types corresponding to these dimensions can be found in the tag library. For example, if there is a dimension in the demand feature vector representing "topic", the system will find the best matching tag under the "topic tag" type in the tag library. The feature value of each dimension (such as a value between 0 and 1) will be compared with the tag in the tag library, and the most suitable tag will be determined by calculating the similarity or matching degree. Once all dimensions are mapped in the tag library, the system will output a set of tags with the highest matching degree, which represent the features that best meet the user's needs. These tags will be used in subsequent recommendation, retrieval or feedback mechanisms to provide users with the most relevant resources or information. For example, in a corpus recommendation system, the user enters the demand information "I need the latest review article on machine learning", and the generated demand feature vector can be [0.9 (topic dimension), 0.8 (time dimension), 0.7 (source dimension)]. Then in the "topic" tag type, the feature vector with a matching degree of 0.9 is mapped to the "machine learning" tag. In the "time" tag type, the feature vector with a matching degree of 0.8 is mapped to the "last year" tag. In the “Source” label type, the feature vector with a match of 0.7 is mapped to the “Top Conferences” label.

[0072] S103, according to a preset demand matching formula, calculating the matching degree between the demand feature vector and each label under each label type in a preset label library; wherein the preset demand matching formula is:

[0073]

[0074] Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B kThe conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product of , which calculates the similarity between the two vectors. The larger the dot product, the higher the matching degree between the demand feature vector and the label. The matching degree can be used to measure the overall similarity between the demand feature vector and the labels in the label library. The larger the value, the higher the matching degree.

[0075] Key features can be extracted from user demand information, historical interaction data, and current context information, and these features can be quantified into dimensions k in the demand feature vector. For example, in the corpus demand matching scenario, the dimensions can include topics, language styles, emotional expressions, and text types. The tag library contains multiple tags, each of which has its own feature vector in each dimension.

[0076] The modulus calculation represents the feature strength in each dimension. For example, the modulus of the demand feature is It can represent the overall strength or weight of the demand feature vector. Using the transformation matrix Transform or standardize the label feature vector to obtain a new feature vector

[0077] The dot product represents the similarity between the demand feature vector and the converted label feature vector in each dimension. The numerator represents the strength of the demand feature and the label feature. The denominator represents the similarity between the demand feature and the label feature after conversion. By dividing, the product of the modulus length is compared with the dot product to measure the matching degree between the demand and the label in each dimension. Multiply the matching degree component of each dimension by the corresponding weight w k, and then add the results of all dimensions to get the final matching degree. And according to this method, the matching degree of the demand feature vector and all tags under each tag type in the preset tag library is calculated. For example, the user needs to extract suitable corpus from the database to support specific research or application. The demand feature vector of the corpus is represented by multiple attributes (such as subject, language style, emotional expression, text type, etc.). The determined demand feature vector is the user demand information: the user currently needs a "formal" text in the field of "science and technology" and hopes that the text is "logically rigorous". Historical interaction data: the corpus that the user used in the past tends to be "neutral emotion" and "clear structure". Current situation information: the user is preparing to write a technical report, which is suitable for citing "high-precision" corpus. After quantifying the features, the demand feature vector is [0.9, 0.8, 0.7, 0.6], corresponding to the following dimensions: Dimension 1: subject (0.9); Dimension 2: language style (0.8); Dimension 3: emotional expression (0.7) Dimension 4: text type (0.6). The corpus tag library in the system contains multiple tags, and each corpus has feature values ​​in different dimensions. The feature vector of corpus 1 is [0.85, 0.75, 0.8, 0.65]. For dimension 1: First, calculate the modulus length of the required feature is 0.9, and then calculate the label feature modulus of corpus 1 is 0.85, using the transformation matrix Transform the topic tag features of corpus 1 to obtain Calculate the dot product of the required features and the converted label features Finally, calculate the matching degree component:

[0078]

[0079] The calculation method for other dimensions is the same as that for dimension 1. Then the matching degree components of each dimension are multiplied by the corresponding weights and the sum is added to obtain the corresponding matching degree.

[0080] S104: If there is a match degree exceeding the preset match degree threshold, the tag corresponding to the match degree exceeding the preset match degree threshold is used as the target tag, response information is generated according to the target tag, and the response information is fed back to the user.

[0081] The preset matching degree threshold may be a minimum matching degree standard set to determine whether the matching degree between the demand feature vector and the tag is strong enough. If the calculated matching degree is higher than the threshold, the tag is considered to be highly relevant to the user demand.

[0082] The target tags may refer to those tags whose matching degree with the required feature vector exceeds the preset matching degree threshold. These tags represent the corpus or content features that are most relevant to the user's needs. By matching these tags, the system can find the corpus resources that best meet the user's needs.

[0083] The response information can be a recommendation or feedback information generated by the system based on the target tag. This information can be a specific response to the user's needs, including relevant recommendations, materials, or further operation suggestions. The response information is intended to provide content or resources that meet the user's needs and help the user quickly find the required corpus or information. For example, the user enters the following demand in the corpus demand matching scenario: Find "review articles on the latest technologies in artificial intelligence" The user's historical interaction data is the user's previous reading of review articles related to artificial intelligence. Current scenario information: The user is now writing a research paper on artificial intelligence. The system generates response information based on the user's demand feature vector and target tags, such as "artificial intelligence" (technical field), "review articles" (research type), "within the last three months" (publication time), such as "We have found some review articles on the latest technologies in artificial intelligence for you. They are all the latest research published in the last three months and are suitable for your reference when writing research papers. Please check these recommended articles to obtain the latest technical trends and research results." This response information provides users with specific recommended materials and matches the user's demand characteristics to help users find the most relevant corpus resources.

[0084] The matching degree between the demand feature vector and each tag in the tag library can be calculated by the demand matching formula. Then the calculated matching degree is compared with the preset matching degree threshold. All tags whose matching degree exceeds the preset threshold are regarded as target tags. According to the target tag, the response information related to the user's demand is generated and sent to the user through wireless communication technology. For example, the user enters the demand in the corpus demand matching system: "Need to find the latest review article on 'Development Trend of Artificial Intelligence'". First, it is converted into a demand feature vector: Demand information: Find a review article on 'Development Trend of Artificial Intelligence' Demand feature vector: The feature vector generated based on the demand information can include dimensions such as "Topic: Artificial Intelligence", "Document Type: Review", "Time Range: Last Six Months". Then calculate the matching degree between the demand feature vector and each tag in the tag library. There are tags in the tag library: such as "Artificial Intelligence", "Technical Review", "Recent Release" Then use the demand matching formula to calculate the matching degree between the demand feature vector and each tag in the tag library. For example, the matching degree of the "Artificial Intelligence" tag may be 0.8, the matching degree of the "Technical Review" tag may be 0.75, and the matching degree of the "Recent Release" tag may be 0.9. The match degree is then compared with the preset match degree threshold. If the preset match degree threshold is 0.7, the labels whose match degree exceeds the threshold are: "Artificial Intelligence" match degree 0.8 (exceeds the threshold), "Technical Review" match degree 0.75 (exceeds the threshold), "Recent Release" match degree 0.9 (exceeds the threshold), and then a response message is generated, for example, "We have found some of the latest review articles on 'Artificial Intelligence Development Trends' for you. These articles have been published in the past six months and meet your needs." Finally, it is sent to the user.

[0085] It is not difficult to find that, compared with the related art, in the solution provided by the embodiment of the present application, if user demand information is received, user historical interaction data and current scenario information are obtained, the demand feature vector and the dimension of each demand feature vector are determined according to the user demand information, user historical interaction data and current scenario information; the dimension of each demand feature vector is mapped in the preset tag library to determine the tag type corresponding to the dimension of each demand feature vector; according to the preset demand matching formula, the matching degree of each tag under each tag type in the preset tag library is calculated; if there is a matching degree exceeding the preset matching degree threshold, the tag corresponding to the matching degree exceeding the preset matching degree threshold is used as the target tag, and the response information is generated according to the target tag, and the response information is fed back to the user. Through the above-mentioned multi-dimensional adaptive corpus demand matching method, the recommendation obtained by the user can be more accurate, in line with personal habits and current needs, thereby improving satisfaction. As user behavior and environment change, the recommended content can be adjusted to maintain the timeliness and relevance of the recommendation. At the same time, the matching process is efficient, the most suitable tag type can be quickly found, the search efficiency is improved, and the target tag that best meets the user's needs is automatically screened out, which can improve the accuracy of the recommendation results.

[0086] Based on the above technical solution, optionally, after feeding back the response information to the user, the method further includes:

[0087] Determine a similar user group in a preset user database according to the behavior vector;

[0088] Acquire recent behavior information of similar user groups, determine potential recommendation information according to the behavior vector and the recent behavior information, and feed back the potential recommendation information to the user.

[0089] In this solution, the preset user database can be a collection of data including detailed information, historical behavior, preferences, interaction records, etc. of all users. The database is used to analyze user behavior and needs, and help make personalized recommendations and optimize the system.

[0090] A similar user group can refer to a collection of users who have similar behavior patterns, preferences, or needs as the target user in a specific dimension.

[0091] Recent behavior information refers to the activity data of similar user groups in the recent period of time, such as the corpus they recently searched for, the research fields they are interested in, their ratings and comments, etc.

[0092] Potential recommendation information can be personalized recommendation content generated based on recent behavior information of similar user groups, and can include relevant research papers, corpora, research reports or other resources related to user needs.

[0093] The behavior data of the target user can be extracted from the preset user database, the similarity between the behavior vector of the target user and the behavior vectors of other users can be calculated, and the users can be divided into similar user groups according to the similarity. The latest behavior data of similar user groups can be extracted from the preset user database, and these data can be analyzed to understand the current user interests and needs. Based on the behavior information of similar users, the recommendation algorithm is used to generate recommended content that meets the needs of the target user. The generated recommendation information is fed back to the target user to provide personalized recommendation services. For example, user A frequently searches and downloads literature on "machine learning" and "deep learning". The system extracts the behavior data of user A from the user database, generates its behavior vector, and calculates the similarity with the behavior vectors of other users. The system finds that user B and user C have similar interests in the field of "machine learning", and they have recently searched and downloaded literature on "latest algorithm research". Based on the latest behavior of these similar users, the system generates a recommendation list, such as "latest literature on algorithm research" or "latest progress in machine learning". These recommendations will be fed back to user A to provide the latest corpus information that matches his needs.

[0094] In this solution, by comparing with the data of similar user groups, we can more accurately predict user preferences and provide more relevant recommendations.

[0095] Based on the above technical solution, optionally, after feeding back the response information to the user, the method further includes:

[0096] Monitor in real time whether the preset tag library cleanup time has been reached. If the preset tag library cleanup time has been reached, obtain the usage rate of each tag in the preset tag library.

[0097] The tag whose usage rate is lower than a preset tag usage rate threshold is used as a target cleanup tag, and the target cleanup tag is deleted from a preset tag library.

[0098] In this solution, the preset tag library cleanup time can refer to the time interval set by the system, which is used to regularly evaluate the effectiveness and update needs of the tag library. This time point can be set based on actual needs, such as monthly, quarterly, etc.

[0099] Usage rate refers to the frequency at which a tag is actually used within a specified period of time. It can be used to measure the activity or relevance of a tag. For example, if a tag has been used 10 times in the past month, its usage rate is 10 times.

[0100] The preset tag usage rate threshold can be a set minimum usage frequency standard to determine whether a tag needs to be cleaned up. For example, if the threshold is set to 5 times, a tag with a usage frequency lower than this value will be regarded as an infrequently used tag.

[0101] The target cleanup tags may be those tags whose usage rate is lower than a preset tag usage rate threshold. These tags are marked as objects that need to be deleted from the tag library.

[0102] You can continuously track the current time to detect whether the preset tag library cleanup time has been reached. If the cleanup time has been reached, the system triggers the cleanup process. Count the usage frequency of each tag in the preset tag library within the set time period. For example, you can calculate the number of times each tag is used by analyzing the system logs or user interaction records. Compare the usage frequency of all tags with the preset tag usage rate threshold, and those tags with usage frequencies lower than the threshold are the target cleanup tags. Delete all items marked as target cleanup tags from the preset tag library. This step can be performed through batch operations or automated scripts to ensure the update and maintenance of the tag library.

[0103] In this solution, reducing redundant tags in the tag library can ensure that the system is more efficient in processing and matching tags. This can shorten response time and improve user experience.

[0104] Second embodiment

[0105] The second embodiment of the present application relates to a multi-dimensional adaptive corpus demand matching method. The second embodiment is an improvement on the first embodiment, and the second embodiment is substantially the same as the first embodiment, with the main difference being that it specifically includes the following steps:

[0106] S301: If user demand information is received, user historical interaction data and current context information are obtained, and demand feature vectors and dimensions of each demand feature vector are determined according to the user demand information, user historical interaction data and current context information.

[0107] S302 : Mapping is performed in a preset label library according to the dimension of each demand feature vector to determine the label type corresponding to the dimension of each demand feature vector.

[0108] S303, according to a preset demand matching formula, calculating the matching degree between the demand feature vector and each tag under each tag type in a preset tag library; wherein the preset demand matching formula is:

[0109]

[0110] Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w kis the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product calculates the similarity between the two vectors. The larger the dot product, the higher the match between the demand feature vector and the label.

[0111] S304: If there is no matching degree exceeding the preset matching degree threshold, a demand adjustment request is sent to the user.

[0112] A request for adjustment can be a prompt sent by the system to the user to obtain more information or correct the user's needs so that the system can find matching tags or make more accurate recommendations. For example, if the user searches for "latest research on deep learning" and the system cannot find an exact match, the system may send the following request for adjustment: "We did not find any 'latest research on deep learning' that fully meets your needs. Please provide more details or adjust your requirements, such as specifying a specific research field or related topic, and we will do our best to find suitable resources for you."

[0113] If the system detects that it cannot find a tag with a matching degree exceeding a preset threshold, it can generate and send a demand adjustment request to the user, asking whether the demand parameters can be adjusted.

[0114] S305: Receive demand adjustment information sent by the user, and determine whether the demand adjustment information is complete.

[0115] The demand adjustment information can be a user's response to the adjustment request issued by the system, which can be new demand information provided by the user or a correction to the existing demand. It helps the system better understand the user's expectations so as to update the demand feature vector and make new matches. For example, after receiving the demand adjustment request, the user provides the following information: "I can accept research in related fields of natural language processing, not limited to the latest papers. If there are some classic research or related review articles, I am also interested."

[0116] The user receives the requirement adjustment request and provides new information. After the system receives this information, it can verify whether the information is complete, that is, check whether all necessary requirement dimensions are met.

[0117] S306: If the demand adjustment information is complete, update the user demand information according to the demand adjustment information.

[0118] If the demand adjustment information is complete, the system can integrate this information into the user's needs.

[0119] S307, obtaining user historical interaction data and current context information, and determining a demand feature vector according to the updated user demand information, user historical interaction data and current context information.

[0120] The demand feature vector can be regenerated based on the updated demand information, user historical interaction data and current context information.

[0121] S308, calculating the matching degree between the demand feature vector and each tag in the preset tag library according to a preset demand matching formula.

[0122] The preset demand matching formula can be used to calculate the matching degree between the updated demand feature vector and each label in the label library.

[0123] S309: If there is still no matching degree exceeding the preset matching degree threshold, determine a potential tag according to the demand feature vector, and add the potential tag to a preset tag library.

[0124] Potential tags can refer to tags that may be matched after the user demand information is updated, and the system identifies them by analyzing the matching degree between the demand feature vector and the tags in the existing tag library. These tags may not be in the current tag library, but have the potential to become effective tags.

[0125] If there is still no tag whose matching degree exceeds the preset threshold, the matching of the demand feature vector and the existing tags can be analyzed to identify potential tags. The identified potential tags are added to the preset tag library to expand the coverage of the tag library and improve future matching results. For example, in the personalized corpus recommendation scenario, the demand adjustment information is the user's voice input "search for action movies with high-intensity action scenes and directed by well-known action movie directors". If the matching degree of the updated demand feature vector and each tag in the preset tag library is calculated according to the preset demand matching formula, if there is still no matching degree exceeding the preset matching threshold, the system detects that the tag library lacks relevant tags, such as "high-intensity action" or "well-known director". Then it can be added to the tag library as a potential tag.

[0126] It is not difficult to find that in the embodiment of the present application, by continuously adjusting the requirements and analyzing the unmet requirements, new tags can be discovered and introduced to enrich the tag library. Adding potential tags to the tag library helps improve the performance of the system when matching similar requirements in the future and improves the overall recommendation quality.

[0127] Based on the above technical solution, optionally, after adding the potential tags to the preset tag library, the method further includes:

[0128] regenerate response information according to the potential tags, and feed back the response information to the user;

[0129] Sending opinion collection information to users, so that users can provide feedback on their satisfaction with the response information generated after adding tags based on the opinion collection information;

[0130] If satisfactory evaluation information is received from the user, the potential tags are retained in the preset tag library;

[0131] Accordingly, after sending the opinion collection information to the user, the method further includes:

[0132] If unsatisfactory evaluation information is received from the user, the potential tag is deleted from the preset tag library.

[0133] In this solution, the opinion collection information can be a survey or form for obtaining user feedback on the latest generated response information. Its purpose is to understand the user's satisfaction with the added potential tags, so as to decide whether to keep these tags in the tag library. It can include a satisfaction survey, which can include simple rating information (such as 1-5 points) or multiple-choice questions (such as "very satisfied", "satisfied", "average", "unsatisfied", "very unsatisfied"). Open questions: allow users to provide detailed opinions and suggestions (such as "please describe your satisfaction or dissatisfaction with the recommended movie").

[0134] Satisfied evaluation information can be feedback from users expressing satisfaction with the response information after adding potential tags. It can be a high score or a positive evaluation. For example, users give a score within the range of 1-5 or 1-10, and usually 3 points or more (or a specific threshold) indicates satisfaction. The text feedback provided by the user contains positive descriptions or recognition (for example, "The recommended movie is very suitable for my taste, very good!") is satisfactory evaluation information.

[0135] Unsatisfactory evaluation information can be feedback from users expressing dissatisfaction with the response information after adding potential tags. It can be a low score or a negative evaluation. For example, a user gives a low score in the range of 1-5 or 1-10, and usually a score below 3 (or a specific threshold) indicates dissatisfaction. If the text feedback provided by the user contains criticism or negation (for example, "The recommended movie does not meet my expectations and I am a little disappointed."), it is unsatisfactory evaluation information.

[0136] Potential tags can be identified from user demand information. These tags are generated based on the user's demand feature vector, but are not predefined in the tag library. These potential tags are temporarily added to the tag library for testing and verification. Based on the potential tags, new response information is generated. For example, in the corpus demand matching scenario, movies, products or services that meet the new tags can be recommended. The new recommendation results are then fed back to the user as response information. If the potential tag is "modern action movies", the system may generate a new movie recommendation list containing movies that meet the "modern action movies" tag. Then create a form or questionnaire containing survey questions. For example, ask the user about their satisfaction with the new recommendation results, using a rating system and open questions. Send the designed opinion collection information to the user through email, messaging system or in-app notification. The user fills in and submits the questionnaire or feedback form. If the user provides a high score or positive comment, it means that he is satisfied with the new recommendation result. If the user provides a low score or negative comment, it means that he is dissatisfied with the new recommendation result. If the user is satisfied with the newly added tags, these tags are officially added to the preset tag library as the basis for future recommendations. If the user is not satisfied with the newly added tags, delete these tags from the tag library to avoid using these tags in future recommendations.

[0137] In this solution, the recommendation system is optimized by continuously collecting and analyzing user feedback so that it can better meet user needs and dynamically adjust the tag library, thereby improving the accuracy of recommendations and user satisfaction.

[0138] Third embodiment

[0139] The third embodiment of the present application relates to a multi-dimensional adaptive corpus demand matching method. The third implementation is an improvement on the first embodiment, and the specific improvement is that it includes the following steps:

[0140] S401: If user demand information is received, user historical interaction data and current context information are obtained, and demand feature vectors and dimensions of each demand feature vector are determined according to the user demand information, user historical interaction data and current context information.

[0141] S402 : Mapping is performed in a preset label library according to the dimension of each demand feature vector to determine the label type corresponding to the dimension of each demand feature vector.

[0142] S403, according to a preset demand matching formula, calculating the matching degree between the demand feature vector and each tag under each tag type in a preset tag library; wherein the preset demand matching formula is:

[0143]

[0144] Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product calculates the similarity between the two vectors. The larger the dot product, the higher the match between the demand feature vector and the label.

[0145] S404: If there is a match degree exceeding the preset match degree threshold, the tag corresponding to the match degree exceeding the preset match degree threshold is used as the target tag, response information is generated according to the target tag, and the response information is fed back to the user.

[0146] S405: Obtain user feedback information, and determine whether a preset tag library needs to be updated according to the user feedback information.

[0147] User feedback information can be the user's evaluation and suggestions on the recommendation results or tags generated by the system. It can include satisfaction rating: the user's satisfaction with the current recommendation or tag (for example, using a rating of 1 to 5). Specific opinions: detailed comments provided by users, such as the pros and cons of the recommendation results, whether the tags match their actual needs, etc. Suggestions: User suggestions on how to improve the tag library, such as adding new tags, modifying the description of existing tags, etc.

[0148] You can collect user comments about the recommended results or tags. Specifically, you can obtain feedback through questionnaires, rating systems, comment boxes, etc. Evaluate user satisfaction scores and analyze users' specific opinions and suggestions to identify tags that need to be updated. Based on user feedback, determine whether the tag library needs to be updated. If the feedback indicates that the existing tags cannot meet the needs or the user suggests a new tag, the tag library needs to be updated.

[0149] S406: If the tag library needs to be updated, update information is determined according to the user feedback information, and the preset tag library is updated according to the update information.

[0150] Update information can refer to the changes made to the tag library based on user feedback, which can include adding new tags: adding new tags based on user needs or suggestions. Modifying tags: adjusting the description or attributes of existing tags to better meet user needs. Deleting tags: removing tags that users consider invalid or irrelevant.

[0151] If the user suggests adding a new tag, define the new tag and its description based on the user's needs or suggestions. Add the new tag and its description to the tag library, and ensure that the system can handle these new tags correctly. For example, if the user suggests adding an "environmentally friendly material" tag, the specific characteristics and application scope of the tag need to be defined. If the user suggests modifying a tag, adjust the description of the existing tag based on the feedback. Update the tag description in the tag library to ensure that the new tag definition takes effect. For example, if the user believes that the description of the "sports shoes" tag is inaccurate, the definition of the tag can be modified. If the user suggests deleting a tag, remove the tags that the user believes are no longer applicable. For example, if user feedback indicates that some outdated tags are useless, these tags can be deleted.

[0152] It should be noted that the third embodiment of the present application may also be an improvement based on the second embodiment.

[0153] It is not difficult to find that in the embodiment of the present application, by updating the tag library according to user feedback, the system can better meet the actual needs of users. This means that the recommendation results will be more accurate and more in line with user expectations. Dynamic adjustment can adapt to changes in market and user needs and improve the flexibility of recommendations.

[0154] Based on the above technical solution, optionally, after updating the preset tag library according to the update information, the method further includes:

[0155] According to a preset demand matching formula, recalculate the matching degree between the demand feature vector and each label under each label type in the preset label library;

[0156] If there is a match degree exceeding the preset match degree threshold, the tag corresponding to the match degree exceeding the preset match degree threshold is used as the target tag, response information is generated according to the target tag, and the response information is fed back to the user;

[0157] Continue to obtain user feedback information, and determine whether to continue updating the preset tag library according to the user feedback information;

[0158] If the preset tag library still needs to be updated, the update information is re-determined according to the user feedback information, and the preset tag library is updated according to the update information, until it is determined according to the user feedback information that the preset tag library does not need to be updated.

[0159] In this solution, a demand feature vector can be determined from user demand information, historical interaction data, and current scenario information. A preset demand matching formula is used to calculate the matching degree between the demand feature vector and each tag in the tag library. The calculated matching degree is compared with a preset matching degree threshold. If there is a matching degree that exceeds the preset matching degree threshold, the corresponding tag is marked as a target tag, and response information based on the target tag is generated. Specific response information is generated according to the target tag, such as recommended content, products, or services. The response information is sent to the user to ensure that the user receives feedback that meets the needs. A request is sent to the user to ask them about their satisfaction with the generated response information, which can be done in the form of questionnaires, ratings, comments, etc. User feedback on the response information is collected, such as satisfactory or unsatisfactory evaluations. Determine whether the tag library needs to be updated based on the user's feedback information. If the feedback indicates that the tags in the tag library do not completely match the user's needs, they need to be updated. Determine the tags that need to be updated based on the user's feedback information, which can include adding new tags, modifying the definition of existing tags, or deleting tags that are no longer relevant. Update the preset tag library based on the determined update information to ensure that the tag library can more accurately reflect user needs and preferences. If the tag library still needs to be updated, repeat the above steps and continue updating and providing feedback until it is determined based on user feedback that the tag library no longer needs further updating.

[0160] In this solution, by recalculating the matching degree and continuously updating the tag library based on user feedback, the system can provide recommendations that better meet user needs. This reduces the probability of users receiving irrelevant recommendations and improves the relevance and accuracy of recommendations.

[0161] The step division of the above methods is only for the purpose of clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this patent; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this patent.

[0162] Fourth embodiment

[0163] The fourth embodiment of the present application relates to a multi-dimensional adaptive corpus demand matching system, such as Figure 3 As shown, the system specifically includes the following:

[0164] The demand feature vector determination module 501 is used to obtain the user's historical interaction data and current context information upon receiving the user's demand information, and determine the demand feature vector and the dimension of each demand feature vector according to the user's demand information, the user's historical interaction data and the current context information;

[0165] The label type determination module 502 is used to map the dimension of each demand feature vector in a preset label library to determine the label type corresponding to the dimension of each demand feature vector;

[0166] The matching degree calculation module 503 is used to calculate the matching degree between the demand feature vector and each tag under each tag type in the preset tag library according to a preset demand matching formula; wherein the preset demand matching formula is:

[0167]

[0168] Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product calculates the similarity between the two vectors. The larger the dot product, the higher the match between the demand feature vector and the label.

[0169] Feedback module 504 is used to, if there is a match degree exceeding a preset match degree threshold, use the tag corresponding to the match degree exceeding the preset match degree threshold as the target tag, generate response information according to the target tag, and feed back the response information to the user. It is not difficult to find that this embodiment is a system embodiment corresponding to the first embodiment, and this embodiment can be implemented in conjunction with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and in order to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied in the first embodiment.

[0170] It is worth mentioning that all modules involved in this embodiment are logic modules. In practical applications, a logic unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application, but this does not mean that there are no other units in this embodiment.

[0171] The flow chart or block diagram in the accompanying drawings shows the possible architecture, function and operation of the equipment, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated system for hardware that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0172] The scope of the present application is defined by the appended claims rather than the above description, and it is intended that all changes that fall within the meaning and scope of the equivalent elements of the claims are included in the present application. Any figure mark in the claims should not be regarded as limiting the claims involved. In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the device claim may also be implemented by one unit or device through software or hardware. The words "first", "second", etc. are only used to distinguish the description, and do not indicate any particular order, nor can they be understood as indicating or implying relative importance.

[0173] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily mention changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims, and the above embodiments should be regarded as exemplary and non-restrictive.

Claims

1. A multi-dimensional adaptive corpus demand matching method, characterized in that: The method comprises: If user demand information is received, historical user interaction data and current context information are obtained, and demand feature vectors and dimensions of each demand feature vector are determined according to the user demand information, historical user interaction data and current context information; Mapping is performed in a preset label library according to the dimension of each demand feature vector to determine the label type corresponding to the dimension of each demand feature vector; According to the preset demand matching formula, the matching degree between the demand feature vector and each label under each label type in the preset label library is calculated; wherein the preset demand matching formula is: Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product of the two vectors is calculated. The larger the dot product, the higher the matching degree between the demand feature vector and the label. If there is a match degree exceeding the preset match degree threshold, the tag corresponding to the match degree exceeding the preset match degree threshold is used as the target tag, response information is generated according to the target tag, and the response information is fed back to the user.

2. The multi-dimensional adaptive corpus demand matching method according to claim 1, characterized in that: Before obtaining the user's historical interaction data and current context information, the method further includes: Determining whether the user demand information is complete; If the user demand information is incomplete, a request for additional information is sent to the user; Receive the supplementary demand information sent by the user, and determine whether the supplementary demand information is complete; If the demand supplementary information is complete, updating the user demand information according to the demand supplementary information; Accordingly, obtaining the user's historical interaction data and current context information, and determining the demand feature vector according to the user's demand information, the user's historical interaction data and the current context information, includes: The demand feature vector is determined based on the updated user demand information, user historical interaction data and current context information.

3. The multi-dimensional adaptive corpus demand matching method according to claim 1, characterized in that: After calculating the matching degree between the demand feature vector and each tag under each tag type in the preset tag library according to the preset demand matching formula, the method further includes: If there is no match that exceeds the preset match threshold, a demand adjustment request is sent to the user; Receive demand adjustment information sent by the user, and determine whether the demand adjustment information is complete; If the demand adjustment information is complete, updating the user demand information according to the demand adjustment information; Obtaining user historical interaction data and current context information, and determining a demand feature vector based on updated user demand information, user historical interaction data, and current context information; Calculate the matching degree between the demand feature vector and each label in the preset label library according to the preset demand matching formula; If there is still no match degree exceeding the preset match degree threshold, a potential tag is determined according to the demand feature vector, and the potential tag is added to a preset tag library.

4. The multi-dimensional adaptive corpus demand matching method according to claim 3, characterized in that: After adding the potential tags to the preset tag library, the method further includes: regenerate response information according to the potential tags, and feed back the response information to the user; Sending opinion collection information to users, so that users can provide feedback on their satisfaction with the response information generated after adding tags based on the opinion collection information; If satisfactory evaluation information is received from the user, the potential tags are retained in the preset tag library; Accordingly, after sending the opinion collection information to the user, the method further includes: If unsatisfactory evaluation information is received from the user, the potential tag is deleted from the preset tag library.

5. The multi-dimensional adaptive corpus demand matching method according to claim 1, characterized in that: After feeding back the response information to the user, the method further includes: Obtain user feedback information, and determine whether a preset tag library needs to be updated according to the user feedback information; If the tag library needs to be updated, update information is determined according to the user feedback information, and the preset tag library is updated according to the update information.

6. The multi-dimensional adaptive corpus demand matching method according to claim 5, characterized in that: After updating the preset tag library according to the update information, the method further includes: According to a preset demand matching formula, recalculate the matching degree between the demand feature vector and each label under each label type in the preset label library; If there is a match degree exceeding the preset match degree threshold, the tag corresponding to the match degree exceeding the preset match degree threshold is used as the target tag, response information is generated according to the target tag, and the response information is fed back to the user; Continue to obtain user feedback information, and determine whether to continue updating the preset tag library according to the user feedback information; If the preset tag library still needs to be updated, the update information is re-determined according to the user feedback information, and the preset tag library is updated according to the update information, until it is determined according to the user feedback information that the preset tag library does not need to be updated.

7. The multi-dimensional adaptive corpus demand matching method according to claim 1, characterized in that: If user demand information is received, historical user interaction data and current context information are obtained, and a demand feature vector is determined according to the user demand information, historical user interaction data and current context information, including: Determine the semantic information, intent information and keyword information of user demand information; Determine an initial feature vector based on the semantic information, intention information and keyword information; Matching the target historical behavior in the user historical interaction data according to the semantic information, the intention information, and the keyword information, and determining a behavior vector of the target historical behavior; A scenario vector is determined according to current scenario information, and a demand feature vector is determined according to the scenario vector, the initial feature vector and the behavior vector.

8. The multi-dimensional adaptive corpus demand matching method according to claim 7, characterized in that: After feeding back the response information to the user, the method further includes: Determine a similar user group in a preset user database according to the behavior vector; Acquire recent behavior information of similar user groups, determine potential recommendation information according to the behavior vector and the recent behavior information, and feed back the potential recommendation information to the user.

9. The multi-dimensional adaptive corpus demand matching method according to claim 1, characterized in that: After feeding back the response information to the user, the method further includes: Monitor in real time whether the preset tag library cleanup time has been reached. If the preset tag library cleanup time has been reached, obtain the usage rate of each tag in the preset tag library. The tag whose usage rate is lower than a preset tag usage rate threshold is used as a target cleanup tag, and the target cleanup tag is deleted from a preset tag library.

10. A multi-dimensional adaptive corpus demand matching system, characterized in that: The system comprises: A demand feature vector determination module is used to, if receiving user demand information, obtain user historical interaction data and current context information, and determine demand feature vectors and dimensions of each demand feature vector according to the user demand information, user historical interaction data and current context information; A label type determination module is used to map the dimension of each demand feature vector in a preset label library to determine the label type corresponding to the dimension of each demand feature vector; The matching degree calculation module is used to calculate the matching degree between the demand feature vector and each label under each label type in the preset label library according to the preset demand matching formula; wherein the preset demand matching formula is: Among them, M f is the matching degree between the demand feature vector u and the jth label in the label library; k is the index used to identify the specific dimension; K is the total number of dimensions of the demand feature vector and each label in the label library; w k is the weight of the kth dimension. The kth dimension refers to the representation of the demand feature vector and the label library features in different aspects. When the demand feature vector and the labels in the label library are associated in multiple feature dimensions, k represents one of these feature dimensions. is the feature vector of the demand feature vector u in the kth dimension. This vector captures the specific characteristics of user demand in a specific dimension; for The module length; For the jth tag in the tag library in field B k A feature vector that describes the attributes or characteristics of a label in a specific domain; for The module length; From field A k To area B k The conversion matrix is ​​used to convert the requirement feature vector from one domain to another domain in cross-domain matching to achieve cross-domain requirement mapping and label matching; is the demand feature vector and the transformed label feature vector The dot product of the two vectors is calculated. The larger the dot product, the higher the matching degree between the demand feature vector and the label. The feedback module is used to, if there is a matching degree exceeding a preset matching degree threshold, use the tag corresponding to the matching degree exceeding the preset matching degree threshold as the target tag, generate response information according to the target tag, and feed back the response information to the user.

Images