Information tracking method and device for expert portrait

Abstract

The present application provides an information tracking method and device for expert portraits, the method comprising: based on the basic expert data of the target expert in the expert information database and the resume data of the target expert, a large language model is used to obtain a missing information query formula for the resume data; the missing information query formula is used to find the missing information-related web pages in the network and identify the target web page text data; the target web page text data and document data are extracted, aggregated and verified for the missing data used to represent the missing information, and the newly added expert attribute description phrases and their semantic information are verified, and then the label level of the newly added expert attribute description phrases is determined based on the expert review rules. The present application can improve the comprehensiveness and accuracy of information tracking for expert portraits; and can improve the efficiency, accuracy and reliability of information tracking for expert portraits, thereby improving the real-time, accuracy and comprehensiveness of updating or constructing expert portraits.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an information tracking method and device for expert portraits. Background Art

[0002] Expert portraits are designed to outline the full picture of experts from relevant data, and they have broad application prospects in research fields such as expert feature analysis, expert recommendation, and expert evaluation. At present, although intelligent information technology has been applied in expert review in fields such as electricity, the following problems still exist: First, the level of technical application is not high, and many companies still rely on traditional methods, such as resumes and forms filled out by experts themselves; second, the effect of technical application is not good, and data from various regions, departments, and systems are not interoperable. There are problems such as untimely tracking of expert information and incomplete characterization, which will also lead to a long information tracking process for expert portraits and increase labor costs; third, the existing expert portrait system does not fully perform semantic analysis of multi-source data for the identification of expert labels, resulting in inaccurate automatic extraction of expert information and the inability to guarantee the accuracy of expert portrait data processing.

[0003] Therefore, there is an urgent need to design a method that can improve the comprehensiveness of information tracking for expert portraits, improve the efficiency of information tracking for expert portraits, and improve the accuracy of tracking data processing. Summary of the invention

[0004] In view of this, the embodiments of the present application provide an information tracking method and device for expert portraits to eliminate or improve one or more defects existing in the prior art.

[0005] One aspect of the present application provides an information tracking method for expert portraits, comprising:

[0006] Based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, a large language model is used to obtain a missing information query formula for the resume data; the missing information query formula is used to search for corresponding missing information-related web pages in the network, and target web page text data is identified from the missing information-related web pages;

[0007] Extracting missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert; filling in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and performing integrity verification on the target expert information template, and updating the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification;

[0008] If there is a newly added expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, the newly added expert attribute description phrase is verified for authenticity and the semantic information of the newly added expert attribute description phrase that passes the authenticity verification is obtained, and the newly added expert attribute description phrase and the corresponding semantic information are stored in the attribute library;

[0009] The tag type to which the newly added expert attribute description phrase belongs is obtained, and based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase and the tag type, the tag level of the newly added expert attribute description phrase is obtained and stored.

[0010] In some embodiments of the present application, based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, a large language model is used to obtain a missing information query formula for the resume data, including:

[0011] Receiving a review request for a target expert, wherein the review request includes current resume data of the target expert;

[0012] Searching for basic expert data of the target expert from the expert information database;

[0013] Inputting the basic expert data and the resume data of the target expert into a large language model, so that the large language model searches for missing information in the resume data compared to the basic expert data and generates a target query formula for the missing information based on a preset prompt word for searching for missing information and generating a query formula, wherein the target query formula includes: a network query formula;

[0014] The network query formula is synonymously expanded based on a preset synonym word set to obtain a missing information query formula for the resume data.

[0015] In some embodiments of the present application, the review request also includes document data corresponding to the target expert, and the target query formula also includes: a document query formula;

[0016] Correspondingly, the application of the missing information query formula to search for corresponding missing information-related web pages in the network, and identifying target web page text data from the missing information-related web pages, includes:

[0017] Applying the missing information query formula to search for corresponding missing information-related web pages in the network, and identifying web page text data from the missing information-related web pages;

[0018] If the number of the missing information-associated web pages found is multiple, the document query is synonym-expanded based on a preset synonym word set, and based on the document query after the synonym expansion, the corresponding document text data is extracted from the document data corresponding to the target expert and / or the pre-stored target expert basic document data;

[0019] Based on the document text data, the accuracy of the web page text data corresponding to each of the missing information associated web pages is verified, and the website confidence corresponding to each of the missing information associated web pages is determined according to the accuracy verification results corresponding to each of the missing information associated web pages, and the web page text data corresponding to the missing information associated web page with the highest website confidence is selected as the current target web page text data.

[0020] In some embodiments of the present application, extracting missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert includes:

[0021] Determine the expert attribute identification type to which the missing information corresponding to the resume data belongs, wherein the expert attribute identification type includes: at least one of an extraction type, a statistical type, and a semantic reasoning type;

[0022] If the expert attribute identification type to which the missing information belongs includes the extraction class, an extraction prompt word corresponding to the missing information belonging to the extraction class is set, and the extraction prompt word, the target web page text data, and the document data corresponding to the target expert are input into the large language model, so that the large language model extracts the missing data corresponding to the missing information belonging to the extraction class from the target web page text data and the document data corresponding to the target expert based on the extraction prompt word.

[0023] In some embodiments of the present application, the integrity verification of the target expert information template and updating the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification include:

[0024] Inputting a preset integrity verification prompt word and the target expert information template into a large language model, so that the large language model performs integrity verification on the target expert information template based on the integrity verification prompt word and outputs corresponding integrity verification result data;

[0025] Determine whether the target expert information template passes the integrity verification based on the integrity verification result data; if so, update the basic expert data corresponding to the target expert in the expert information database based on the target expert information template, wherein the expert information database is used to store the basic expert data of each expert, and each of the basic expert data is formed based on a preset expert information template.

[0026] In some embodiments of the present application, if there is a newly added expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, the newly added expert attribute description phrase is authenticated and the semantic information of the newly added expert attribute description phrase that passes the authenticity verification is obtained, and the newly added expert attribute description phrase and the corresponding semantic information are stored in the attribute library, including:

[0027] Performing data denoising processing on the target expert information template;

[0028] If there is a newly added expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, the target website corresponding to the newly added expert attribute description phrase is retrieved from the network, and the authenticity of the newly added expert attribute description phrase is verified from the webpage content in the target website;

[0029] If the newly added expert attribute description phrase passes the authenticity verification, the semantic information of the newly added expert attribute description phrase is extracted from the webpage content in the target website, and the newly added expert attribute description phrase and the corresponding semantic information are stored in a preset attribute library.

[0030] In some embodiments of the present application, the step of obtaining and storing the tag level of the newly added expert attribute description phrase based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase, and the tag type includes:

[0031] Inputting the semantic information and label type of the newly added expert attribute description phrase into a large language model pre-constructed based on preset expert review rule data, so that the large language model outputs the label level of the newly added expert attribute description phrase accordingly;

[0032] The corresponding relationship between the newly added expert attribute description phrase and the label level of the newly added expert attribute description phrase is stored in a preset expert attribute library.

[0033] Another aspect of the present application provides an information tracking device for expert portraits, comprising:

[0034] A query formula generation and network query module is used to obtain a missing information query formula for the resume data based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert using a large language model; the missing information query formula is used to search for the corresponding missing information related web pages in the network, and the target web page text data is identified from the missing information related web pages;

[0035] An information extraction, aggregation and verification module is used to extract the missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert; fill in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and perform integrity verification on the target expert information template, and update the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification;

[0036] A newly added attribute verification and storage module is used to verify the authenticity of the newly added expert attribute description phrase and obtain the semantic information of the newly added expert attribute description phrase that passes the authenticity verification if there is a newly added expert attribute description phrase that is not currently stored in the preset attribute library in the target expert information template, and store the newly added expert attribute description phrase and the corresponding semantic information in the attribute library;

[0037] The tag classification and rating module is used to obtain the tag type to which the newly added expert attribute description phrase belongs, and based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase and the tag type, obtain the tag level of the newly added expert attribute description phrase and store it.

[0038] The third aspect of the present application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the information tracking method for expert portraits when executing the computer program.

[0039] The fourth aspect of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the information tracking method for expert portraits.

[0040] The fifth aspect of the present application provides a computer program product, including a computer program, which implements the information tracking method for expert portraits when executed by a processor.

[0041] The information tracking method for expert portraits provided by the present application is based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, and adopts a large language model to obtain a missing information query formula for the resume data; applies the missing information query formula to search for the corresponding missing information-related web pages in the network, and identifies the target web page text data from the missing information-related web pages; extracts the missing data corresponding to the resume data from the target web page text data and the document data corresponding to the target expert; fills in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and performs integrity verification on the target expert information template, and updates the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification; if there is a target expert information template that is not currently stored For the newly added expert attribute description phrases stored in the preset attribute library, the authenticity of the newly added expert attribute description phrases is verified and the semantic information of the newly added expert attribute description phrases that pass the authenticity verification is obtained, and the newly added expert attribute description phrases and the corresponding semantic information are stored in the attribute library; the tag type to which the newly added expert attribute description phrases belong is obtained, and based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrases and the tag type, the tag level of the newly added expert attribute description phrases is obtained and stored. By adopting multi-source data such as expert information library, expert resume, web page data, expert files and expert review rule data, the comprehensiveness and accuracy of information tracking for expert portraits can be effectively improved; by adopting a large language model, the efficiency, accuracy and reliability of information tracking for expert portraits can be effectively improved, thereby improving the real-time, accuracy and comprehensiveness of updating or constructing expert portraits.

[0042] Additional advantages, purposes, and features of the present application will be partially described in the following description, and will become partially apparent to those skilled in the art after studying the following, or may be learned from the practice of the present application. The purposes and other advantages of the present application can be achieved and obtained by the structures specifically pointed out in the specification and the drawings.

[0043] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present application are not limited to the above specific description, and the above and other purposes that can be achieved by the present application will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The drawings described herein are used to provide a further understanding of the present application, constitute a part of the present application, and do not constitute a limitation of the present application. The components in the drawings are not drawn to scale, but are only for the purpose of illustrating the principles of the present application. In order to facilitate the illustration and description of some parts of the present application, the corresponding parts in the drawings may be enlarged, that is, they may become larger relative to other components in the exemplary device actually manufactured according to the present application. In the drawings:

[0045] Figure 1 A flowchart of an information tracking method for expert portraits in one embodiment of the present application.

[0046] Figure 2 1 is a flowchart of step 100 in the information tracking method for expert portraits in one embodiment of the present application.

[0047] Figure 3 It is a flowchart of step 200 in the information tracking method for expert portraits in one embodiment of the present application.

[0048] Figure 4 Schematic diagram of the process of step 300 in the information tracking method for expert portraits in one embodiment of the present application.

[0049] Figure 5 It is a flowchart of step 400 in the information tracking method for expert portraits in one embodiment of the present application.

[0050] Figure 6 It is a structural diagram of an information tracking device for expert portraits in one embodiment of the present application.

[0051] Figure 7 This is a functional architecture diagram of an information tracking method for expert portraits in an application example of the present application.

[0052] Figure 8 The figure is a schematic diagram showing an example of the process of specificity verification and Web search enhancement in an application example of the present application. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the implementation modes and the accompanying drawings. Here, the illustrative implementation modes and descriptions of the present application are used to explain the present application, but are not intended to limit the present application.

[0054] It should also be noted here that in order to avoid obscuring the present application due to unnecessary details, only the structures and / or processing steps closely related to the scheme according to the present application are shown in the accompanying drawings, while other details that are not very relevant to the present application are omitted.

[0055] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0056] It should also be noted that, unless otherwise specified, the term “connection” herein may refer not only to a direct connection but also to an indirect connection involving an intermediate.

[0057] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0058] In order to design a method that can improve the comprehensiveness of information tracking for expert portraits, improve the efficiency of information tracking for expert portraits, and improve the accuracy of tracking data processing, the embodiments of the present application respectively provide an information tracking method for expert portraits, an information tracking device for expert portraits for executing the information tracking method for expert portraits, an electronic device, a computer-readable storage medium, and a computer program product, which can improve the comprehensiveness and accuracy of information tracking for expert portraits; and can improve the efficiency, accuracy, and reliability of information tracking for expert portraits, can effectively save the time and labor costs of expert portraits, and thus can improve the real-time, accuracy, and comprehensiveness of updating or constructing expert portraits.

[0059] The details are described in detail through the following examples.

[0060] Based on this, the embodiment of the present application provides an information tracking method for expert portraits that can be implemented by an information tracking device for expert portraits, see Figure 1 The information tracking method for expert portraits specifically includes the following contents:

[0061] Step 100: Based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, a large language model is used to obtain a missing information query formula for the resume data; the missing information query formula is used to search the network for corresponding missing information-related web pages, and the target web page text data is identified from the missing information-related web pages.

[0062] The expert information database refers to a database stored locally by an enterprise or university, etc., which is used to store the basic data of enterprise experts. The basic data of enterprise experts contains the corresponding relationship between the expert unique identification (such as ID, etc.) corresponding to each expert and the basic expert data. It can be understood that the basic expert data is used to represent the complete historical information of the expert and can be generated according to the expert information template.

[0063] Among them, the expert information database can be set up as one or more according to actual application requirements. In one example, the expert information database can adopt an industry benchmark expert database and an industry knowledge database.

[0064] In one or more embodiments of the present application, the large language model (LLM) can directly adopt an open source large language model, or it can be used after fine-tuning the open source large language model. Among them, the architecture of the large language model can adopt the Llama3 model or the ChatGLM model, etc., and the fine-tuning can adopt the XTuner large model fine-tuning toolkit, which can realize a variety of fine-tuning training such as QLoRA, LoRA and full parameter fine-tuning. The choice of fine-tuning method is to choose full fine-tuning when the hardware computing power is sufficient, and to choose QLoRA or LoRA when the computing power is limited. The fine-tuning data uses the expert-related information existing in the open source or enterprise memory, and constructs training samples through data, including entity recognition, question-answer pairs, etc. In addition, information tracking for expert portraits can be achieved based on LLM's retrieval enhanced generation (Retrieval Augmented Generation), fine-grained prompt word construction, task decomposition, multi-step inference and LLM fine-tuning.

[0065] Step 200: Extract the missing data corresponding to the resume data from the target web page text data and the document data corresponding to the target expert; fill in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and perform integrity verification on the target expert information template; and update the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification.

[0066] In step 200, the missing data corresponding to the resume data refers to the data corresponding to the information missing from the resume data compared to the basic expert data of the target expert.

[0067] Step 300: If there is a new expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, the authenticity of the new expert attribute description phrase is verified and the semantic information of the new expert attribute description phrase that passes the authenticity verification is obtained, and the new expert attribute description phrase and the corresponding semantic information are stored in the attribute library.

[0068] Step 400: Obtain the tag type to which the newly added expert attribute description phrase belongs, and based on preset expert review rule data, semantic information of the newly added expert attribute description phrase and the tag type, obtain and store the tag level of the newly added expert attribute description phrase.

[0069] From the above description, it can be seen that the information tracking method for expert portraits provided in the embodiment of the present application can effectively improve the comprehensiveness and accuracy of information tracking for expert portraits by adopting multi-source data such as expert information database, expert resume, web page data, expert files and expert review rule data; by adopting a large language model, it can effectively improve the efficiency, accuracy and reliability of information tracking for expert portraits, and can effectively save the time and manpower costs of expert portraits, thereby improving the real-time, accuracy and comprehensiveness of expert portrait updating or construction.

[0070] In order to further improve the effectiveness and efficiency of generating missing information query formulas for the resume data, and to further improve the effectiveness and efficiency of information tracking for expert portraits, in an information tracking method for expert portraits provided in an embodiment of the present application, see Figure 2 , step 100 in the information tracking method for expert portraits specifically includes the following contents:

[0071] Step 110: Receive a review request for a target expert, wherein the review request includes current resume data of the target expert.

[0072] Step 120: Searching for basic expert data of the target expert from the expert information database.

[0073] Step 130: Input the basic expert data and the resume data of the target expert into a large language model, so that the large language model searches for missing information in the resume data compared to the basic expert data and generates a target query formula for the missing information based on preset prompt words for searching for missing information and generating query formulas, wherein the target query formula includes: a network query formula.

[0074] Step 140: Synonym expansion is performed on the network query expression based on a preset synonym word set to obtain a missing information query expression for the resume data.

[0075] In order to further improve the effectiveness and reliability of searching for web pages associated with missing information, and to further improve the accuracy and reliability of information tracking for expert portraits, in an information tracking method for expert portraits provided in an embodiment of the present application, the review request also includes the document data corresponding to the target expert, and the target query formula also includes: document query formula; see Figure 2 The information tracking method for expert portraits further specifically includes the following contents after step 140 in step 100:

[0076] Step 150: Use the missing information query formula to search for corresponding missing information-related web pages in the network, and identify web page text data from the missing information-related web pages.

[0077] Step 160: If the number of web pages associated with the missing information found is multiple, the document query formula is synonym expanded based on a preset synonym word set, and based on the document query formula after the synonym expansion, the corresponding document text data is extracted from the document data corresponding to the target expert and / or the pre-stored target expert basic document data.

[0078] Step 170: Based on the document text data, the accuracy of the web page text data corresponding to each of the missing information associated web pages is verified, and the website confidence corresponding to each of the missing information associated web pages is determined according to the accuracy verification results corresponding to each of the missing information associated web pages, and the web page text data corresponding to the missing information associated web page with the highest website confidence is selected as the current target web page text data.

[0079] Specifically, the information tracking device for expert portraits can automatically generate missing information queries required for information updates by comparing the complete basic expert data and the resume data of the current target expert. The implementation method is to design and train a query generation model based on LLM. LLM is used in conjunction with query generation prompts, that is, prompts for finding missing information and generating query types. After reading the basic expert data and resume data of the current target expert to be analyzed, the missing content in the resume data of the current target expert is identified.

[0080] In one example, the time when the expert received the award is missing. For the search source type, the missing information query generated includes:

[0081] (1) The web query type contains data for indicating the expert's name, the expert's unit or institution, and the expert's information known in the resume data such as the scientific and technological progress award; that is, the web page must contain the name of the person and the unit. In the queried information page, it is identified whether the corresponding time information is included.

[0082] (2) Document query: contains data for representing the expert's name, the expert's unit or institution, scientific and technological progress award, and the known expert information in the resume data, so as to extract similar text fragments of the document. Identify whether the corresponding time information is contained in the queried document.

[0083] In order to further improve the effectiveness and reliability of extracting missing data corresponding to resume data, and to further improve the accuracy and reliability of information tracking for expert portraits, in an information tracking method for expert portraits provided in an embodiment of the present application, see Figure 3 , step 200 in the information tracking method for expert portraits specifically includes the following contents:

[0084] Step 210: Determine the expert attribute identification type to which the missing information corresponding to the resume data belongs, wherein the expert attribute identification type includes: at least one of an extraction type, a statistical type, and a semantic reasoning type.

[0085] Step 220: If the expert attribute identification type to which the missing information belongs includes the extraction class, an extraction prompt word corresponding to the missing information belonging to the extraction class is set, and the extraction prompt word, the target web page text data, and the document data corresponding to the target expert are input into the large language model, so that the large language model extracts the missing data corresponding to the missing information belonging to the extraction class from the target web page text data and the document data corresponding to the target expert based on the extraction prompt word.

[0086] Specifically, expert attribute recognition types can be divided into three types: extraction, statistics, and semantic reasoning. The extraction type mainly includes achievements and awards received, which can only be extracted from the original document; the statistical type labels are labels for age, number of papers, etc., which need to be calculated and counted based on specific information in the document; the semantic reasoning type requires setting the reasoning logic, such as "management expert", "research expert", "technical expert", etc., which may not appear in the original document, but can be semantically recognized through the document content.

[0087] In addition, if the expert attribute identification type to which the missing information belongs includes the statistical class, the target web page text data and the document data corresponding to the target expert are input into a preset statistical classification model, so that the statistical classification model obtains the missing data corresponding to the missing information belonging to the statistical class from the target web page text data and the document data corresponding to the target expert, wherein the document data corresponding to the target expert includes: pre-stored target expert basic document data and / or document data corresponding to the target expert extracted from the received review request for the target expert.

[0088] Step 230: Fill in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert.

[0089] In one or more embodiments of the present application, the expert information template (prompt) may specify the type of data to be extracted, including: basic information, achievements (time, achievement name), awards (time, achievement), position (time, position), etc. Table 1 is an example of the data type in an expert information template (prompt).

[0090] Table 1

[0091]

[0092]

[0093] In order to further improve the effectiveness and reliability of the integrity verification of the target expert information template, and to further improve the accuracy and reliability of the information tracking for the expert portrait, in an information tracking method for the expert portrait provided in an embodiment of the present application, see Figure 3 The information tracking method for expert portraits further specifically includes the following contents after step 230 in step 200:

[0094] Step 240: input the preset integrity verification prompt word and the target expert information template into the large language model, so that the large language model performs integrity verification on the target expert information template based on the integrity verification prompt word and outputs corresponding integrity verification result data.

[0095] Step 250: Determine whether the target expert information template passes the integrity verification based on the integrity verification result data; if so, update the basic expert data corresponding to the target expert in the expert information database based on the target expert information template, wherein the expert information database is used to store the basic expert data of each expert, and each of the basic expert data is formed based on a preset expert information template.

[0096] Specifically, the acquisition of expert information needs to go through multiple steps of verification, and multiple verification prompts can be set to verify the integrity of the extracted information. Set prompts to instruct LLM to confirm whether the original text mentions a certain label and provide label details. For example, ask LLM whether the document mentioned clearly states that a certain expert has a certain award, award level, award time and issuing department, and verify the results with the existing expert portrait labels.

[0097] In order to solve the dynamic change of expert portraits and adapt to the emerging new phrases required for expert attribute description, in an information tracking method for expert portraits provided in an embodiment of the present application, see Figure 4 Step 300 of the information tracking method for expert portraits specifically includes the following contents:

[0098] Step 310: performing data denoising processing on the target expert information template.

[0099] Step 320: If the target expert information template contains a new expert attribute description phrase that is not currently stored in the preset attribute library, the target website corresponding to the new expert attribute description phrase is retrieved from the network, and the authenticity of the new expert attribute description phrase is verified from the web page content in the target website.

[0100] Step 330: If the newly added expert attribute description phrase passes the authenticity verification, the semantic information of the newly added expert attribute description phrase is extracted from the webpage content in the target website, and the newly added expert attribute description phrase and the corresponding semantic information are stored in a preset attribute library.

[0101] In order to obtain the label level of the newly added expert attribute description phrase, in an information tracking method for expert portraits provided in an embodiment of the present application, see Figure 5 , step 400 in the information tracking method for expert portraits specifically includes the following contents:

[0102] Step 410: Determine the label type to which the newly added expert attribute description phrase belongs, and input the semantic information and label type of the newly added expert attribute description phrase into a large language model pre-constructed based on preset expert review rule data, so that the large language model outputs the label level of the newly added expert attribute description phrase accordingly.

[0103] Step 420: storing the correspondence between the newly added expert attribute description phrase and the label level of the newly added expert attribute description phrase in a preset expert attribute library.

[0104] Specifically, this application proposes to identify the quantity, level, regional scope, and other quantifiable contents contained in the label to obtain a semantic label with metricity. The data based on it is mainly the relevant content in various review documents. First, by extracting documents about institutions, journals, honors, awards, etc. in the literature, extract paragraphs containing label sequences in the documents, identify the "better than" relationship between labels through the text arrangement order, and count the votes according to the sorting order mentioned in multiple documents to obtain the order relationship between labels.

[0105] After obtaining the order relationship of similar labels, the expert automatic evaluation can calculate the quantitative value of the expert label set through the entropy weight method, and summarize the number, level, regional scope and time of different types of labels. The number mainly refers to the number of papers and achievements, the level refers to the level of publications or awards, and the regional scope refers to the level of the title or award-issuing institution, such as national or provincial and ministerial levels. In addition, the academician title and the Outstanding Youth Fund can be comprehensively calculated and assigned according to the time of acquisition, the order of ranking, the number of participants and the number of winners.

[0106] The hierarchical correspondence between expert labels and their text descriptions forms a semantic hierarchical relationship with metricity. This application represents the semantic information of labels as semantic coding, builds and trains LLM based on the review standard document, and obtains an LLM model with the ability to classify labels. The trained model can effectively automatically extract expert representative works and representative achievements from the expert innovation ability, achievement quality, effectiveness, and contribution reflected in the document details such as undertaking projects, applying for patents, publishing papers, and achievement conversion.

[0107] From the software level, the present application also provides an information tracking device for expert portraits for executing all or part of the information tracking method for expert portraits, see Figure 6 The information tracking device for expert portraits specifically includes the following contents:

[0108] The query formula generation and network query module 10 is used to obtain the missing information query formula for the resume data based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert using a large language model; apply the missing information query formula to search for the corresponding missing information-related web pages in the network, and identify the target web page text data from the missing information-related web pages.

[0109] The information extraction, aggregation and verification module 20 is used to extract the missing data corresponding to the resume data from the target web page text data and the document data corresponding to the target expert; fill in the preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and verify the integrity of the target expert information template; and update the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification.

[0110] The newly added attribute verification and storage module 30 is used to verify the authenticity of the newly added expert attribute description phrase and obtain the semantic information of the newly added expert attribute description phrase that passes the authenticity verification if there is a newly added expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, and store the newly added expert attribute description phrase and the corresponding semantic information in the attribute library.

[0111] The tag classification and rating module 40 is used to obtain the tag type to which the newly added expert attribute description phrase belongs, and based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase and the tag type, obtain the tag level of the newly added expert attribute description phrase and store it.

[0112] The embodiment of the information tracking device for expert portraits provided in the present application can be specifically used to execute the processing flow of the embodiment of the information tracking method for expert portraits in the above-mentioned embodiment. Its functions will not be repeated here, and reference can be made to the detailed description of the embodiment of the information tracking method for expert portraits mentioned above.

[0113] The part of the information tracking device for expert portraits that performs information tracking for expert portraits can be executed in the server or completed in the client device. The specific selection can be based on the processing capability of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor for specific processing of information tracking for expert portraits.

[0114] The client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and other implementation scenarios may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0115] The server and the client device may communicate with each other using any suitable network protocol, including network protocols that have not yet been developed on the date of filing this application. The network protocols may include, for example, TCP / IP, UDP / IP, HTTP, HTTPS, etc. Of course, the network protocols may also include, for example, RPC (Remote Procedure Call Protocol) and REST (Representational State Transfer) protocols used on top of the above protocols.

[0116] From the above description, it can be seen that the information tracking device for expert portraits provided in the embodiment of the present application can effectively improve the comprehensiveness and accuracy of information tracking for expert portraits by adopting multi-source data such as expert information database, expert resume, web page data, expert files and expert review rule data; by adopting a large language model, it can effectively improve the efficiency, accuracy and reliability of information tracking for expert portraits, and can effectively save the time and labor costs of expert portraits, thereby improving the real-time, accuracy and comprehensiveness of expert portrait updating or construction.

[0117] To further illustrate the above embodiment, the present application also provides a specific application example of an information tracking method for expert portraits. This application example is based on the information database of science and technology benchmark experts, detects highly reliable information sources and source characteristics, and then monitors the information sources to obtain multi-source expert data information in real time. Through the label extraction technology based on multi-document data, the typical characteristics of power grid science and technology innovation experts are mined, and expert portrait information is tracked and updated. It has practical value in improving expert identification and management efficiency, supporting decision-making, etc.

[0118] This application example uses heterogeneous information from multiple sources such as Internet information and expert academic achievements to automatically supplement and compare expert data. By integrating multi-source information of experts, a real-time, accurate and comprehensive expert portrait is obtained.

[0119] This application example uses the development method of prompt word engineering and utilizes the Large Language Model (LLM) to realize the automated processing of the expert profiling system, including multi-source information acquisition, analysis, information extraction, and inspection and verification.

[0120] The construction of the labeling system for multi-source data fusion is based on the real-time network information completion of the existing benchmark expert database. It automatically generates queries to crawl the personal description information of benchmark experts in the field, obtains the text description of the benchmark expert information, and automatically crawls the academic and monograph information of benchmark experts to form an expert database. The network information acquisition and processing method here can also be used for the acquisition and processing of new expert information in the future. For the functional architecture of the information tracking method for expert portraits provided in the application example of this application, please refer to Figure 7 ,It can use the current expert tag system to automatically extract expert features, greatly reducing the ,difficulties brought by manual identification.,Specifically, this application example first uses the existing industry benchmark expert ,database and industry knowledge base to dynamically obtain Web search results, supplement expert data in real ,time, and find and aggregate expert tags from them.

[0121] The information tracking method for expert portraits provided in this application example specifically includes the following contents:

[0122] 1. Obtain basic expert data of target experts in the industry through the expert information database, expressed in the form of text labels, such as "Academician", "Outstanding Youth" and "Doctoral Supervisor".

[0123] 2. Use the benchmark expert database to complete the expert data format, compare the missing data in the current resume data of the current target expert, generate a missing information query for the data items to be updated, retrieve Web pages and the company's own text data, and obtain multi-source expert information.

[0124] 3. The characteristics of expert portraits are diverse data sources and high data authenticity requirements. This application example extracts the missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert; fills in the preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and verifies the integrity of the target expert information template, and updates the basic expert data corresponding to the target expert in the expert information database based on the target expert information template that has passed the integrity verification, and calculates the confidence of the data item based on the data source, data release time, and multi-source data mutual verification method.

[0125] 4. Based on the preset expert review rule data (such as guiding documents on the evaluation, training, and development of industry expert talents), the semantic information of the newly added expert attribute description phrases, and the label type, the newly added expert attribute description phrases in the formatted expert resume (i.e., the target expert information template) obtained after the search enhancement are automatically labeled, and the label level of the newly added expert attribute description phrases is obtained, and the expert labels corresponding to the expert list are displayed in different colors for easy identification. For example, types correspond to different colors, and levels correspond to different color saturations.

[0126] Multi-source data refers to data generated from different sources and in multiple forms. By integrating, consolidating, sharing and associating multi-source data, an instant, accurate and comprehensive description can be obtained. Multi-source data is obtained from: expert information database, expert resume, web page data, expert documents and expert review rules data (such as expert review standard documents) and other multi-source data.

[0127] The multi-source information data acquisition process specifically includes the following contents:

[0128] 1. Automatic query generation: The information tracking device for expert portraits can automatically generate missing information queries required for information updates by comparing the complete basic expert data and the resume data of the current target expert. The implementation method is to design and train a query generation model based on LLM. Use LLM with query generation prompt words, that is, prompt words used to find missing information and generate query expressions. After reading the basic expert data and resume data of the current target expert to be analyzed, identify the missing content in the resume data of the current target expert.

[0129] 2. Synonym expansion: By constructing a synonym set, we can retrieve synonyms, replace synonyms one by one in the search formula, search multiple times and summarize the results to improve the recall rate of related web pages. For example, the names of institutions, awards, and achievements all have full names and multiple abbreviations, and the date of birth also has multiple formats.

[0130] 3. Information source tracking: Manually set the reliability of the information source website, that is, through the website information confidence calculation, automatically set a higher confidence for the website that obtains valid information multiple times:

[0131] (1) Assign higher confidence to authoritative websites. This part includes judging the validity of the information obtained from the website.

[0132] (2) Multi-source data verification is adopted, and the documents submitted by the author and the existing data of the enterprise are used as verification data to verify the number of hits and the number of new additions between the website information and the existing information, and high confidence is set for the websites that can provide high-quality expert information.

[0133] (3) The set of high-confidence websites is denoted as Website trusted ={w1, w2, w3, ...}, w1, W2 and W3 refer to the confidence of different websites respectively.

[0134] 4. Web page information acquisition:

[0135] Web crawler tools such as Playwright crawl web pages and download updates.

[0136] Use web content parsing tools to parse HTML web pages, retain the main text content, and identify and convert formatted data content such as tables and lists. Data parsing can be done with the help of tools such as BeautifulSoup, which is a Python package for parsing HTML and XML documents.

[0137] Table and list data are formatted data, usually with high accuracy, and it is easy to extract complete field names and field values ​​with the help of their format. The method is to use web page content parsing tools to parse the tags of tables and lists in web pages.

[0138] 5. Use the comparison between complete data and current data. Check whether the information in the existing document library within the enterprise contains relevant information, and extract the related content in the relevant documents. PDF, Word and other file data recognition can use specific text format recognition tools to extract text information from files, and should also focus on the extraction of table and list data. The recognition method can call advanced multimodal document recognition tools, such as GLM4. The recognized text content is the same as the above web page content. The free text is segmented and sent to LLM to extract attribute tags. Convert the tabular data into formatted data for statistical analysis.

[0139] The expert label extraction, aggregation and verification process of multi-source information specifically includes the following contents:

[0140] 1. Multi-source data extraction and fusion:

[0141] For example, use the schema information extraction function provided by the large language model.

[0142] The basic information in the expert information template (such as name, gender, ethnicity, date of birth and place of origin) is unique and unchanging information and is used to identify the expert. Other information is dynamically changing information and can be updated in real time through user-uploaded documents, network and other information.

[0143] 2. Expert information extraction and verification:

[0144] Retrieval of enhanced formatted expert information is based on the updated page obtained by Web information tracking, from which the attribute tags possessed by the expert are identified. To achieve this function, an expert information extraction module based on LLM is first established. The prompt words are manually set, and the effect of the prompt words is adjusted and verified to obtain the prompt word template.

[0145] Expert attribute identification types can be divided into three types: extraction, statistics, and semantic reasoning. The extraction type mainly includes achievements and awards received, which can only be extracted from the original document; the statistical type labels are age stages, number of papers, etc., which need to be calculated and counted based on specific information in the document; the semantic reasoning type requires setting the reasoning logic, such as "management expert", "research expert" and "technical expert", which may not appear in the original document, but can be semantically identified through the document content.

[0146] The following takes the acquisition of extraction class labels as an example to give the prompt words required for information extraction. The prompt word project is a process of cyclic testing and optimization. The following examples are only used for examples and do not limit the prompt words.

[0147] (1) The information extraction model sends the prompt words and the text segment to be recognized to the LLM, and instructs the LLM model to extract the expert information contained in various source data such as Word, PDF, text paragraphs, Web pages, etc.

[0148] For example, prompt="Extract and save the related entities and their attributes mentioned in the following paragraphs. Only the attributes mentioned in the "information_extraction" function are extracted. If an attribute does not exist and is not required in the function parameters, it is not included in the output. Paragraph: {input}". The prompt word {} is a placeholder and needs to be replaced with a specific string when used.

[0149] (2) Local information completion: By filling in the required information as in the prompt template, you will get the prompt words for local information completion. All types of information are similar and will not be described in detail. Examples are as follows:

[0150] Complete the rewards received by the expert, such as prompt="Extract the following paragraphs that mention {expert name + expert unit} {rewards received}, including but not limited to {list of existing awards}. If {rewards received} does not exist and is not required in the function parameters, it is not included in the output. Paragraph: {input}".

[0151] (3) Statistical labels need to be obtained by calculating and counting specific information in the document. At present, LLM is not good at analyzing and counting data, so a specially designed statistical classification model is needed.

[0152] (4) Semantic reasoning attributes are based on expert data, using simplified expert labels to represent expert features. Based on labels such as expert data research field, expertise, and type, they can be automatically labeled using multi-label classification. Basic attributes and labels that require authorization and certification from a single institution cannot be automatically labeled.

[0153] (5) Experts extract information verification:

[0154] The acquisition of expert information requires multiple steps of verification. Multiple verification prompts can be set to verify the integrity of the extracted information. Set prompts to instruct LLM to confirm whether the original text mentions a certain label and provide label details. For example, ask LLM whether the document mentioned clearly states that a certain expert has a certain award, award level, award time and issuing department, and verify the results with the existing expert portrait labels.

[0155] In addition, in order to solve the dynamic characteristics of expert portraits and adapt to the emerging new phrases required for expert attribute descriptions, such as the "Ho Leung Ho Lee Award" and "Outstanding Young Energy Science and Technology Worker" and other awards or titles. This application example uses the method of expert description phrase retrieval and phrase description text semanticization to add new expert attribute phrases to the expert attribute library. Through the Internet, the candidate tags are accurately retrieved, and the query is set to the candidate tags enclosed in English double quotes. By detecting the context in which the attribute phrase appears, the introduction text about the new tag is obtained. The real existence of its attribute phrase is confirmed through the page content, and the objective description of the attribute phrase is obtained through the associated text introduction.

[0156] Among them, the process of specificity verification and Web retrieval enhancement is exemplified by Figure 8 As shown in the figure, the verification of expert attribute description phrases can be achieved through human-computer collaboration. However, building a new label verification model to assist manual verification can minimize the manual workload and work difficulty. This application example proposes a multi-strategy fusion evaluation and verification method to verify it. The expert attribute phrases extracted by the model must have the properties of universality, authenticity, and objectivity.

[0157] (1) The reliability of the expert attribute phrases that have been confirmed in the system can be calculated through text string matching.

[0158] (2) For phrases that fail to match, there may be three possibilities:

[0159] ① The problem of multiple words with the same meaning in a label. The problem of multiple words with the same meaning can be solved by calculating semantic similarity. The calculation of semantic similarity of text requires semantic encoding of the text first, such as using models such as Sentence-Bert to convert the text into a vector, and then calculating the similarity score by, for example, cosine similarity. The cosine similarity takes the value of [-1,1], and the larger the value, the higher the similarity.

[0160] ② For new information related to expert attributes, retrieve the corresponding text through web pages to obtain the relevant page content of the authoritative website of expert attributes:

[0161] 1) New institutions, colleges, honors, awards, and titles are information that has been published on news release web pages. For example, by searching for "China Electric Power Science and Technology Progress Award", we can find the official website of the awarding institution, find its description page, classify "China Electric Power Science and Technology Progress Award" into "China Electric Power Science and Technology Award", and obtain its description text.

[0162] 2) New professional directions usually already exist and are gradually forming a scale. Therefore, through web page retrieval and semantic matching calculation, the real existence of the tag can be confirmed and its related description can be obtained.

[0163] ③ Words that are not suitable as labels need to be filtered out. Noise labels that are not suitable as expert features appear, or labels are not relevant to the current expert. The reasons are:

[0164] 1) Text that appears in the description section of a resume, header, footer, sidebar, or title may contain phrases that are expert tags, which can be easily misjudged. Therefore, the text content needs to be cleaned and screened before extracting tags.

[0165] 2) The model incorrectly associates attributes when text describing other people appears in the current expert description information. It is necessary to increase the recognition of whether the expert described in the page is a single person or multiple people. In the case of multiple names, it is necessary to distinguish different areas of the expert description.

[0166] (3) The award criteria is also a more specific description of the award selection, which should be added to the associated text of the expert attribute tag. The text is semantically represented and stored to represent the details of the expert attributes.

[0167] The expert label classification and grading process is exemplified as follows:

[0168] This application example proposes to perform semantic expansion on phrase tags, obtain relevant descriptions of tags by performing semantic retrieval on tags, and perform semantic encoding on tags.

[0169] 1. Tag classification

[0170] This application example proposes two tasks for semantic mining of expert tags: tag classification and grading of similar tags.

[0171] When selecting experts for the "A Region Science and Technology Rising Star" award, for example, a similar label set is obtained by comparing the label similarities of the selection guidance documents for the science and technology rising star experts with the selection guidance documents for other awards in the knowledge base.

[0172] Taking a certain management method as an example, the indicator sentence "The Rising Star Program includes two categories: innovative rising stars and entrepreneurial rising stars, mainly for young experts under the age of 35 (relaxed to 37 for women) who are engaged in applied basic research, technological innovation and engineering technology research and development in Region A" is extracted, among which "innovative rising stars" and "entrepreneurial rising stars" are secondary categories of scientific and technological rising stars in Region A; "Region A", "under 35 years old" and "women under 37 years old" are screening conditions; "applied basic research", "technological innovation" and "engineering technology research and development" are expert types. By calculating the semantic similarity of labels, it can be judged that "technical experts" and "engineering experts" in this system are more suitable for participating in the selection than "management experts".

[0173] 2. Expert labeling

[0174] Since expert differentiation mainly requires the grading of labels of the same type, automatic identification of label grades plays an important role in automatic expert grading. For example, type B institutions are better than ordinary type C institutions, type C institutions are better than ordinary type D institutions, and national awards are usually better than provincial and ministerial awards of the same type. This ranking order exists in the guidance documents of various review conditions. This application example proposes to use the review standard document as the expert selection measurement scheme standard and construct an expert measurement model based on the review standard. Label grading mainly relies on the current grading of units, institutions, awards, publications, conferences, etc. at all levels, and then maps them to individual experts with such labels.

[0175] This application example proposes to identify the quantity, level, regional scope, and other quantifiable contents contained in the label to obtain a semantic label with metricity. The data based on it is mainly the relevant content in various review documents. First, by extracting documents about institutions, journals, honors, awards, etc. from the literature, extract paragraphs containing label sequences in the documents, identify the "better than" relationship between labels through the text arrangement order, and count the votes according to the sorting order mentioned in multiple documents to obtain the order relationship between labels.

[0176] After obtaining the order relationship of similar labels, the expert automatic evaluation can calculate the quantitative value of the expert label set through the entropy weight method, and summarize the number, level, regional scope, time and other factors of different types of labels. The number mainly refers to the number of papers and achievements, the level refers to the level of publications or awards, and the regional scope refers to the level of the title or award-issuing institution, such as national, provincial and ministerial levels. In addition, the academician title, outstanding youth fund, etc. can be comprehensively calculated and assigned according to the time of acquisition, ranking order, number of participants, number of winners, etc.

[0177] The hierarchical correspondence between expert labels and their text descriptions forms a semantic hierarchical relationship with metricity. This application example represents the semantic information of labels as semantic coding, builds and trains LLM based on the review standard document, and obtains an LLM model with the ability to classify labels. The trained model can effectively automatically extract expert representative works and representative achievements from the expert innovation ability, achievement quality, effectiveness, and contribution reflected in the document details such as undertaken projects, application example patents, published papers, and achievement conversion.

[0178] In other words, this application example builds a model for acquiring and integrating expert portrait information generated by LLM. Compared with traditional expert portrait systems, this model can be quickly adjusted according to different expert information sources and portrait task requirements. This flexibility enables this system to more effectively handle diverse and growing expert portrait information.

[0179] The multi-source data used in this application example include basic data of enterprise experts, data published in real time on the Web, expert review application case books and review standard documents, etc. The data are rich in form and semantically diverse. This application example uses the text understanding and extraction functions of LLM to build specific functional modules to achieve data processing. The functions include automatic identification of missing data and generation of text query expressions, multi-dimensional feature extraction and verification of experts, and obtaining expert labels using the content of expert evaluation standard documents.

[0180] The embodiment of the present application also provides an electronic device, which may include a processor, a memory, a receiver and a transmitter, wherein the processor is used to execute the information tracking method for expert portraits mentioned in the above embodiment, wherein the processor and the memory may be connected via a bus or other means, such as by bus connection. The receiver may be connected to the processor and the memory via wired or wireless means.

[0181] The processor may be a central processing unit (CPU). The processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or a combination of the above chips.

[0182] As a non-transient computer-readable storage medium, the memory can be used to store non-transient software programs, non-transient computer executable programs and modules, such as program instructions / modules corresponding to the information tracking method for expert portraits in the embodiment of the present application. The processor executes various functional applications and data processing of the processor by running the non-transient software programs, instructions and modules stored in the memory, that is, the information tracking method for expert portraits in the above method embodiment is implemented.

[0183] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the processor, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0184] The one or more modules are stored in the memory, and when executed by the processor, perform the information tracking method for expert portraits in the embodiment.

[0185] In some embodiments of the present application, the user equipment may include a processor, a memory, and a transceiver unit, which may include a receiver and a transmitter. The processor, memory, receiver, and transmitter may be connected through a bus system. The memory is used to store computer instructions, and the processor is used to execute the computer instructions stored in the memory to control the transceiver unit to send and receive signals.

[0186] As an implementation method, the functions of the receiver and the transmitter in the present application can be considered to be implemented through a transceiver circuit or a dedicated chip for transceiver, and the processor can be considered to be implemented through a dedicated processing chip, a processing circuit or a general chip.

[0187] As another implementation method, it is possible to use a general-purpose computer to implement the server provided in the embodiment of the present application, that is, to store the program code for implementing the functions of the processor, receiver, and transmitter in a memory, and the general-purpose processor implements the functions of the processor, receiver, and transmitter by executing the code in the memory.

[0188] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned information tracking method for expert portraits are implemented. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a floppy disk, a hard disk, a removable storage disk, a CD-ROM, or any other form of storage medium known in the technical field.

[0189] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the information tracking method for expert portraits.

[0190] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is specifically performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0191] It should be clear that the present application is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present application.

[0192] In the present application, features described and / or illustrated for one embodiment may be used in the same manner or in a similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0193] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An information tracking method for expert portraits, characterized in that: include: Based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, a large language model is used to obtain the missing information in the resume data compared to the basic expert data and generate a network query formula for the missing information, and the network query formula is synonymously expanded based on a preset synonym word set to obtain a missing information query formula for the resume data; The missing information query formula is applied to search for corresponding missing information-related web pages in the network, and target web page text data is identified from the missing information-related web pages; the network query formula is a target query formula, and the target query formula also includes a document query formula for extracting corresponding document text data from the document data corresponding to the target expert and / or pre-stored target expert basic document data after the synonym expansion when the number of the missing information-related web pages found is multiple; Extracting missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert; filling in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and performing integrity verification on the target expert information template, and updating the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification; If there is a newly added expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, the newly added expert attribute description phrase is verified for authenticity and the semantic information of the newly added expert attribute description phrase that passes the authenticity verification is obtained, and the newly added expert attribute description phrase and the corresponding semantic information are stored in the attribute library; Obtaining the tag type to which the newly added expert attribute description phrase belongs, and based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase and the tag type, obtaining the tag level of the newly added expert attribute description phrase and storing it; The step of extracting missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert includes: Determine the expert attribute identification type to which the missing information corresponding to the resume data belongs; If the expert attribute identification type to which the missing information corresponding to the resume data belongs includes an extraction class, an extraction prompt word corresponding to the missing information belonging to the extraction class is set, and the extraction prompt word, the target webpage text data, and the document data corresponding to the target expert are input into the large language model, so that the large language model extracts the missing data corresponding to the missing information belonging to the extraction class from the target webpage text data and the document data corresponding to the target expert based on the extraction prompt word; If the expert attribute identification type to which the missing information belongs includes a statistical class, the target webpage text data and the document data corresponding to the target expert are input into a preset statistical classification model, so that the statistical classification model obtains the missing data corresponding to the missing information belonging to the statistical class from the target webpage text data and the document data corresponding to the target expert; If the expert attribute identification type to which the missing information belongs includes a semantic reasoning class, the target web page text data and the document data corresponding to the target expert are marked based on various preset tags to obtain the missing data corresponding to the missing information belonging to the semantic reasoning class.

2. The information tracking method for expert portraits according to claim 1 is characterized in that: Based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, a large language model is used to obtain the missing information in the resume data compared to the basic expert data and generate a network query formula for the missing information, including: Receiving a review request for a target expert, wherein the review request includes current resume data of the target expert; Searching for basic expert data of the target expert from the expert information database; The basic expert data and the resume data of the target expert are input into a large language model, so that the large language model searches for missing information in the resume data compared to the basic expert data and generates a target query formula for the missing information based on preset prompt words for searching for missing information and generating query formulas, wherein the target query formula includes: a network query formula.

3. The information tracking method for expert portraits according to claim 2 is characterized in that: The review request also includes document data corresponding to the target expert. Correspondingly, the application of the missing information query formula to search for corresponding missing information-related web pages in the network, and identifying target web page text data from the missing information-related web pages, includes: Applying the missing information query formula to search for corresponding missing information-related web pages in the network, and identifying web page text data from the missing information-related web pages; If the number of the missing information-associated web pages found is multiple, the document query is synonym-expanded based on a preset synonym word set, and based on the document query after the synonym expansion, the corresponding document text data is extracted from the document data corresponding to the target expert and / or the pre-stored target expert basic document data; Based on the document text data, the accuracy of the web page text data corresponding to each of the missing information associated web pages is verified, and the website confidence corresponding to each of the missing information associated web pages is determined according to the accuracy verification results corresponding to each of the missing information associated web pages, and the web page text data corresponding to the missing information associated web page with the highest website confidence is selected as the current target web page text data.

4. The information tracking method for expert portraits according to claim 1 is characterized in that: The step of performing integrity verification on the target expert information template and updating basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification includes: Inputting a preset integrity verification prompt word and the target expert information template into a large language model, so that the large language model performs integrity verification on the target expert information template based on the integrity verification prompt word and outputs corresponding integrity verification result data; Determine whether the target expert information template passes the integrity verification based on the integrity verification result data; if so, update the basic expert data corresponding to the target expert in the expert information database based on the target expert information template, wherein the expert information database is used to store the basic expert data of each expert, and each of the basic expert data is formed based on a preset expert information template.

5. The information tracking method for expert portraits according to claim 1 is characterized in that: If the target expert information template contains a newly added expert attribute description phrase that is not currently stored in the preset attribute library, authenticity verification is performed on the newly added expert attribute description phrase, semantic information of the newly added expert attribute description phrase that passes the authenticity verification is obtained, and the newly added expert attribute description phrase and the corresponding semantic information are stored in the attribute library, including: Performing data denoising processing on the target expert information template; If there is a newly added expert attribute description phrase in the target expert information template that is not currently stored in the preset attribute library, the target website corresponding to the newly added expert attribute description phrase is retrieved from the network, and the authenticity of the newly added expert attribute description phrase is verified from the webpage content in the target website; If the newly added expert attribute description phrase passes the authenticity verification, the semantic information of the newly added expert attribute description phrase is extracted from the webpage content in the target website, and the newly added expert attribute description phrase and the corresponding semantic information are stored in a preset attribute library.

6. The information tracking method for expert portraits according to claim 1 is characterized in that: The step of obtaining and storing the tag level of the newly added expert attribute description phrase based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase, and the tag type includes: Inputting the semantic information and label type of the newly added expert attribute description phrase into a large language model pre-constructed based on preset expert review rule data, so that the large language model outputs the label level of the newly added expert attribute description phrase accordingly; The corresponding relationship between the newly added expert attribute description phrase and the label level of the newly added expert attribute description phrase is stored in a preset expert attribute library.

7. An information tracking device for expert portraits, characterized in that: include: A query formula generation and network query module, which is used to use a large language model to obtain missing information in the resume data compared to the basic expert data based on the basic expert data corresponding to the target expert pre-stored in the expert information database and the current resume data of the target expert, and generate a network query formula for the missing information, and perform synonym expansion on the network query formula based on a preset synonym word set to obtain a missing information query formula for the resume data; The missing information query formula is applied to search for corresponding missing information-related web pages in the network, and target web page text data is identified from the missing information-related web pages; the network query formula is a target query formula, and the target query formula also includes a document query formula for extracting corresponding document text data from the document data corresponding to the target expert and / or pre-stored target expert basic document data after the synonym expansion when the number of the missing information-related web pages found is multiple; An information extraction, aggregation and verification module is used to extract the missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert; fill in a preset expert information template based on the resume data and the missing data to form a target expert information template corresponding to the target expert, and perform integrity verification on the target expert information template, and update the basic expert data corresponding to the target expert in the expert information database according to the target expert information template that has passed the integrity verification; A newly added attribute verification and storage module is used to verify the authenticity of the newly added expert attribute description phrase and obtain the semantic information of the newly added expert attribute description phrase that passes the authenticity verification if there is a newly added expert attribute description phrase that is not currently stored in the preset attribute library in the target expert information template, and store the newly added expert attribute description phrase and the corresponding semantic information in the attribute library; A tag classification and rating module, used to obtain the tag type to which the newly added expert attribute description phrase belongs, and based on the preset expert review rule data, the semantic information of the newly added expert attribute description phrase and the tag type, obtain the tag level of the newly added expert attribute description phrase and store it; The step of extracting missing data corresponding to the resume data from the target webpage text data and the document data corresponding to the target expert includes: Determine the expert attribute identification type to which the missing information corresponding to the resume data belongs; If the expert attribute identification type to which the missing information corresponding to the resume data belongs includes an extraction class, an extraction prompt word corresponding to the missing information belonging to the extraction class is set, and the extraction prompt word, the target webpage text data, and the document data corresponding to the target expert are input into the large language model, so that the large language model extracts the missing data corresponding to the missing information belonging to the extraction class from the target webpage text data and the document data corresponding to the target expert based on the extraction prompt word; If the expert attribute identification type to which the missing information belongs includes a statistical class, the target webpage text data and the document data corresponding to the target expert are input into a preset statistical classification model, so that the statistical classification model obtains the missing data corresponding to the missing information belonging to the statistical class from the target webpage text data and the document data corresponding to the target expert; If the expert attribute identification type to which the missing information belongs includes a semantic reasoning class, the target web page text data and the document data corresponding to the target expert are marked based on various preset tags to obtain the missing data corresponding to the missing information belonging to the semantic reasoning class.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the information tracking method for expert portraits as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the information tracking method for expert portraits as described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the information tracking method for expert portraits as described in any one of claims 1 to 6 is implemented.

Images