Search

CN-122019754-A - Campus service handling method and system based on AI large model

CN122019754ACN 122019754 ACN122019754 ACN 122019754ACN-122019754-A

Abstract

The invention discloses a campus service handling method and system based on an AI large model, which relate to the technical field of artificial intelligent data processing, and the method comprises the steps of obtaining a campus service network search log information set, obtaining a new log set through preprocessing, constructing a term and a document vector, calculating a history query recommendation score and forming a query context group, constructing a word-text matrix by combining a vocabulary, obtaining a semantic matrix after decomposition, calculating the aggregation weight of the term and a candidate term, obtaining a final query by screening and expanding the weight and the new query, outputting a result according to a user habit format by a campus service system, attaching a semantic remark, solving the problem that search efficiency is low due to the fact that query and document semantic gap cannot be made up by ignoring the query context, and effectively improving user satisfaction and accuracy of the query result.

Inventors

  • CHEN YIRONG
  • ZHANG TAOFENG
  • CHEN HAOQIANG

Assignees

  • 正元智慧集团股份有限公司

Dates

Publication Date
20260512
Application Date
20260127

Claims (10)

  1. 1. A campus service handling method based on AI big model, the method comprising: Acquiring a log information set in a campus service officer network search engine, and preprocessing various information in the log information set to obtain a new log information set, wherein the new log information set comprises a new query document and a vocabulary; Constructing a term vector and a document vector according to the new log information set, calculating a recommendation score of the historical query according to the term vector and the document vector, and constructing a query context group through the recommendation score and the historical query document; Constructing a word matrix according to the query context group and the vocabulary, decomposing the word matrix to obtain a semantic matrix, and calculating a condensation weight for semantic vectors of each term and semantic vectors of each candidate term in the semantic matrix; and screening and expanding the new query document through the aggregate weight to obtain a final expanded query, and outputting the final query result according to a user habit format by the campus service system and attaching semantic remarks.
  2. 2. The AI-large-model-based campus service transaction method of claim 1, wherein constructing term vectors and document vectors from the new log information set includes: judging the historical query document and the new query document by taking a vocabulary as a dimension, if the term belongs to the vocabulary, the corresponding position value of the vector is 1, and if the term does not belong to the vocabulary, the corresponding position value of the vector is 0, so as to obtain the term vector of the historical query document and the term vector of the new query document; The method comprises the steps that a historical query document is judged by taking a click document ID set of the historical query as a dimension, if the click document is a click document of the historical query document, a vector corresponding position value is 1, and if the document is not the click document of the historical query document, the vector corresponding position value is 0, so that a document vector of the historical query document is obtained, wherein the click document is a document which is actively clicked and opened by a user; And judging the new query document by taking the test set document ID set as a dimension, taking N top ranked documents after the initial retrieval of the new query document, defining the top ranked documents as the retrieval document, wherein if the document belongs to the retrieval document, the corresponding position value is 1, and if the document does not belong to the retrieval document, the corresponding position value is 0, so as to obtain the document vector of the new query document.
  3. 3. The AI-large-model-based campus service transaction method of claim 1, wherein calculating a recommendation score for the historical query from the term vector and the document vector comprises: Counting the occurrence frequency of terms in queries for a new query document and a historical query document, and calculating probability scores of terms in a language model of the historical query document and a language model of the new query document, wherein the probability scores comprise the probability of the language model of the historical query document and the probability of the language model of the new query document; And calculating the similarity of the terms according to the probability score, and normalizing and outputting the recommendation score of each historical query document relative to the new query document.
  4. 4. The AI-large-model-based campus service transaction method of claim 2 wherein constructing a word matrix from the query context groups and the vocabulary includes: Arranging all the historical query documents according to a recommendation score descending order to obtain a mapping relation between the historical query documents and the recommendation score, determining the number of optimal recommendation query documents through the mapping relation, and screening out the first K historical query documents as recommendation query documents; obtaining a query context group by collecting new query documents, retrieving documents and recommending query document combinations; Defining all term times of the occurrence of the query context group as row dimensions, defining the number of documents contained in the query context group as column dimensions after +1, and constructing a blank matrix according to the row dimensions and the column dimensions; the method comprises the steps of calculating the weight of terms in a real document by using TF-IDF as a first weight, regarding a new query document as a virtual document, and calculating the weight of each term in the new query document as a second weight, wherein the real document is a document combination of a retrieval document and a recommended query document after de-duplication; and filling the first weight and the second weight to the corresponding positions of the empty matrix to obtain the word matrix.
  5. 5. The AI-large-model-based campus service transaction method of claim 4 wherein the final expanded query is obtained by aggregating weights and new query text filtering and expanding, comprising: decomposing the word and text matrix according to linear algebra to obtain a term semantic matrix, a singular value diagonal matrix and a document semantic matrix, and performing truncation and dimension reduction processing on the term semantic matrix, the singular value diagonal matrix and the document semantic matrix to obtain a dimension reduction word and text matrix; Extracting semantic vectors of all the candidate terms of the query context group from the term semantic matrix to serve as first semantic vectors, and extracting semantic vectors of all the candidate terms of the query context group to serve as second semantic vectors; obtaining the aggregation weight of the candidate terms and the new query document according to the first semantic vector and the second semantic vector, arranging all the candidate terms in a descending order according to the aggregation weight, and eliminating the terms repeated with the new query document terms; Screening the first M terms from a multi-source text set to serve as expansion terms according to the preset expansion term number M, wherein the multi-source text set comprises texts of new query documents, search document texts, texts of recommended query documents and click document texts of recommended query documents; The expanded terms are added to the text of the new query document to form the final expanded query.
  6. 6. The campus service handling system based on the AI large model is characterized by comprising a preprocessing module, a construction module, a matrix generation module and an expansion module, wherein: The preprocessing module is used for acquiring a log information set in a campus service officer network search engine, preprocessing various information in the log information set to obtain a new log information set, wherein the new log information set comprises a new query document and a vocabulary; The construction module is used for constructing a term vector and a document vector according to the new log information set, calculating the recommendation score of the historical query according to the term vector and the document vector, and constructing a query context group through the recommendation score and the historical query document; The weight calculation module is used for constructing a word matrix according to the query context group and the vocabulary, decomposing the word matrix to obtain a semantic matrix, and calculating the aggregation weight for the semantic vector of each term and the semantic vector of each candidate term in the semantic matrix; the expansion module is used for screening and expanding the new query document through the aggregation weight to obtain a final expansion query, and the campus service system outputs the final query result according to the habit format of the user and carries with semantic remarks.
  7. 7. The AI-large-model-based campus services handling system of claim 6, wherein the preprocessing module includes a term vector module, a first document vector module, and a second document vector module, wherein: The term vector module is used for judging the historical query document and the new query document by taking the vocabulary as a dimension, if the term belongs to the vocabulary, the corresponding position value of the vector is 1, and if the term does not belong to the vocabulary, the corresponding position value of the vector is 0, so that the term vector of the historical query document and the term vector of the new query document are obtained; The first document vector module is used for judging historical query documents by taking a click document ID set of the historical query as a dimension, if the click document is a click document of the historical query document, the vector corresponding position value is 1, and if the document is not the click document of the historical query document, the vector corresponding position value is 0, so as to obtain a document vector of the historical query document; The second document vector module is used for judging the new query document by taking the test set document ID set as a dimension, taking N top ranked documents after the initial retrieval of the new query document, defining the top ranked documents as the retrieval document, and obtaining the document vector of the new query document if the document belongs to the retrieval document, the corresponding position value is 1, and if the document does not belong to the retrieval document, the corresponding position value is 0.
  8. 8. The AI-large-model-based campus service transaction system of claim 6, wherein the building module includes a probability score calculation module and a recommendation score calculation module, wherein: The probability score calculation module is used for counting the occurrence frequency of terms in the query of the new query document and the historical query document and calculating the probability score of the terms in the language model of the historical query document and the language model of the new query document; And the recommendation score calculating module is used for calculating the similarity of the terms according to the probability score, and carrying out normalization to output the recommendation score of each historical query document relative to the new query document.
  9. 9. The AI-large-model-based campus service transaction system of claim 7, wherein the matrix generation module includes a mapping module, a combining module, a dimension module, a weight calculation and filling module, wherein: The mapping module is used for arranging all the historical query documents according to the recommendation score descending order to obtain the mapping relation between the historical query documents and the recommendation score, determining the number of the optimal recommendation query documents according to the mapping relation, and screening out the first K historical query documents as recommendation query documents; The combination module is used for obtaining a query context group by collecting new query documents, retrieving documents and recommending query document combinations; the dimension module is used for defining all term times of the occurrence of the query context group as row dimensions, defining the number of documents contained in the query context group as column dimensions after +1, and constructing a blank matrix according to the row dimensions and the column dimensions; the weight calculation is used for calculating the weight of the term in the real document by using the TF-IDF as a first weight, treating the new query document as a virtual document and calculating the weight of each term in the new query document as a second weight, wherein the real document is a document combination of the retrieval document and the recommended query document after the weight is removed; the filling module is used for filling the first weight and the second weight to the corresponding positions of the empty matrix to obtain the word matrix.
  10. 10. The AI-large-model-based campus service handling system of claim 9, wherein the expansion module comprises a dimension reduction module, a semantic vector module, a condensation weight module, a screening module, and an addition module, wherein: The dimension reduction module is used for decomposing the word and text matrix according to linear algebra to obtain a term semantic matrix, a singular value diagonal matrix and a document semantic matrix, and carrying out truncation dimension reduction processing on the term semantic matrix, the singular value diagonal matrix and the document semantic matrix to obtain a dimension reduction word and text matrix; The semantic vector module is used for extracting semantic vectors of all the candidate terms of the query context group from the term semantic matrix to serve as a first semantic vector, and extracting semantic vectors of all the candidate terms of the query context group to serve as a second semantic vector; the aggregation weight module is used for obtaining aggregation weights of candidate terms and new query documents according to the first semantic vector and the second semantic vector, arranging all the candidate terms in a descending order according to the aggregation weights, and eliminating terms which are repeated with the terms of the new query documents; The screening module is used for screening the first M terms from a multi-source text set as extension terms according to the preset extension term number M, wherein the multi-source text set comprises a text of a new query document, a search document text, a text of a recommended query document and a click document text of the recommended query document; The adding module is used for adding the expansion terms to the text of the new query document to form a final expansion query.

Description

Campus service handling method and system based on AI large model Technical Field The invention belongs to the technical field of artificial intelligence data processing, and particularly relates to a campus service handling method and system based on an AI large model. Background At present, more and more universities are required to realize full coverage and data of one-number-one-source through one-network, and the universities cross departments and cooperative barriers are broken through by digital tools, so that intelligent campus marker post construction is connected, and services are promoted to be updated from function integration to experience optimization. The technical layer is that campus informatization is from a centralized portal to an intelligent center, based on basic capabilities such as unified identity authentication and flow engines, data asset management is newly added to standardized cleaning and labeled management of educational administration, personnel and financial data, and intelligent enabling in scene is based on teacher and student behaviors to push customized services. Meanwhile, micro-application rapid iteration is realized by means of a low-code platform, and the large AI model supports semantic retrieval and intelligent service guiding. The prior publication number is CN118467679A, which discloses a system and a method for inquiring text big data, comprising the steps of acquiring text big data semantics, determining text big data keywords, performing type analysis on the text big data keywords, determining text big data types and text big data inquiring data, acquiring feedback of a user on the text big data inquiring data based on the text big data inquiring data, inquiring and controlling the text big data inquiring data based on the feedback condition of the user, and intelligently controlling the text big data inquiring data. Although the method improves the efficiency of inquiring the text big data through intelligent control, the problem that the searching effect is poor due to the fact that the inquiring context is ignored, for example, a computer in a repairing laboratory is needed in a campus, a user does not use an independent service entrance such as standard classification information equipment report repair (computer class), the system can return to office furniture repair, hydropower repair and the like, the condition that the searching efficiency is reduced due to ambiguity of the semantics is caused when the similar condition occurs in campus service handling, and the normal operation of campus service is seriously influenced, so that unnecessary economic loss is caused. Disclosure of Invention The invention aims to solve the problem that search efficiency is low because query context is ignored and query and document semantic gaps cannot be made up in campus service handling, and provides a campus service handling method and system based on an AI large model. In a first aspect of the present invention, a campus service handling method based on an AI large model is first provided, where the method includes: Acquiring a log information set in a campus service officer network search engine, and preprocessing various information in the log information set to obtain a new log information set, wherein the new log information set comprises a new query document and a vocabulary; Constructing a term vector and a document vector according to the new log information set, calculating a recommendation score of the historical query according to the term vector and the document vector, and constructing a query context group through the recommendation score and the historical query document; Constructing a word matrix according to the query context group and the vocabulary, decomposing the word matrix to obtain a semantic matrix, and calculating a condensation weight for semantic vectors of each term and semantic vectors of each candidate term in the semantic matrix; and screening and expanding the new query document through the aggregate weight to obtain a final expanded query, and outputting the final query result according to a user habit format by the campus service system and attaching semantic remarks. Optionally, constructing the term vector and the document vector from the new log information set includes: judging the historical query document and the new query document by taking a vocabulary as a dimension, if the term belongs to the vocabulary, the corresponding position value of the vector is 1, and if the term does not belong to the vocabulary, the corresponding position value of the vector is 0, so as to obtain the term vector of the historical query document and the term vector of the new query document; The method comprises the steps that a historical query document is judged by taking a click document ID set of the historical query as a dimension, if the click document is a click document of the historical query document, a vector corresponding position v