Retrieval system and method incorporating a knowledge graph

Abstract

The application provides a retrieval system introducing a knowledge graph, comprising: an interaction module, used for acquiring query information input by a user end; a data storage module, used for storing information data of experts by using ElasticSearch; a recall module, used for carrying out word segmentation on the query information to obtain a list of words, expanding the words according to the knowledge graph to obtain an expanded list of words, calculating the weight of each word according to the expanded list of words, sorting and filtering the words according to the weight, and retrieving according to the filtered words to obtain information of a preset number of experts; a sorting module, used for sorting the preset number of experts and sending information of the most relevant expert to the interaction module according to the quantity of user demand; and the interaction module is further used for providing the information of the most relevant expert to the user end. By introducing the knowledge graph, the application constructs a multi-path recall and realizes accurate retrieval, thereby improving the retrieval effect and having very strong scalability.

Description

Technical Field

[0001] This application relates to the field of retrieval system technology, and in particular to a retrieval system and method that incorporates a knowledge graph. Background Technology

[0002] Traditional search engines are powerful tools for solving information overload, but they suffer from significant shortcomings in accuracy, personalization, and interactive feedback, especially when searching for talent in a specific niche, where the experience is even worse. For Aminer, a website that allows users to search for experts, with its 130 million experts, expert search is an extremely complex task, requiring consideration of both algorithm accuracy and speed, as well as existing data and the daily influx of new data.

[0003] However, users often lack a clear understanding of the information they want to search for, and don't know how to accurately describe what they need. They can only provide a general, broad, and often vague search term. In this case, traditional search engines struggle to determine the user's true search intent from this vague term, thus failing to provide satisfactory results. Summary of the Invention

[0004] This application aims to at least partially address one of the technical problems in the related art.

[0005] Therefore, the first objective of this application is to propose a retrieval system that incorporates a knowledge graph, which solves the technical problem that existing search engines have difficulty in accurately determining the user's true query intent and returning query results. It uses Elasticsearch for storage and msmarco-distilbert-cos-v5 and GBrank technologies for sorting. At the same time, when constructing the query, a knowledge graph is introduced to build multi-path recall and achieve accurate retrieval, thereby improving the retrieval effect and having strong scalability.

[0006] The second objective of this application is to propose a retrieval method that incorporates knowledge graphs.

[0007] The third objective of this application is to propose a computer device.

[0008] The fourth objective of this application is to provide a non-transitory computer-readable storage medium.

[0009] To achieve the above objectives, the first aspect of this application proposes a retrieval system incorporating a knowledge graph, comprising: an interaction module for acquiring query information input by a user; a data storage module for storing expert information data using ElasticSearch; a recall module for segmenting the query information acquired by the interaction module to obtain a list of words, expanding the words according to the knowledge graph to obtain an expanded list of words, calculating the weight of each word according to the expanded list of words, sorting and filtering the words according to the weights, and retrieving information from a preset number of experts based on the filtered words; a sorting module for sorting the preset number of experts obtained by the recall module, and sending the information of the most relevant experts to the interaction module according to the user's demand; the interaction module is also used to provide the information of the most relevant experts to the user.

[0010] Optionally, in one embodiment of this application, the data storage module is specifically used for:

[0011] For long fields in expert information data, ik_smart is used for word segmentation, and then the data is inserted into ElasticSearch for storage.

[0012] For short fields in expert information data, use keywords to insert them into Elasticsearch for storage.

[0013] Optionally, in one embodiment of this application, the query information includes a title and a request text, and the recall module is specifically used for:

[0014] Use NLTK to segment the title and requirement text to obtain a list of words in the title and requirement text;

[0015] The vocabulary is expanded based on the word list and the subject knowledge graph, resulting in an expanded word list, and the source of each word in the list is marked.

[0016] The similarity between each word and the title and demand text is calculated using TF-IDF as the weight of each word;

[0017] The weights of words sourced from the demand text are calculated and updated using the BERT model, and the weights of words sourced from the title are calculated and updated using the DSSM model.

[0018] The weights of words whose sources are labeled as knowledge graph extensions are reduced, and the weights are updated based on the reduced weights.

[0019] The words with updated weights are merged and sorted, and words with weights below a preset threshold are filtered and deleted.

[0020] Searching based on the filtered words yields a preset number of expert profiles.

[0021] Optionally, in one embodiment of this application, the query information further includes keywords and a sorting module, specifically used for:

[0022] The relevance of the title and a preset number of experts is calculated using a relevance calculation model.

[0023] Get the relevance of a preset number of experts returned by Elasticsearch;

[0024] By determining whether the keywords entered by the user match the keywords of a preset number of experts, the keyword matching value of the preset number of experts is obtained.

[0025] The model relevance, ElasticSearch relevance, keyword matching value, h-index, citation count, and paper publication count of a preset number of experts are input into the GBrank model for calculation to obtain the weights of the preset number of experts.

[0026] Based on keyword matching values, the weight of a preset number of experts is increased;

[0027] The experts are sorted according to their weighted values, and the information of the expert with the highest weight is sent to the interaction module according to the number of users' needs.

[0028] Optionally, in one embodiment of this application, it further includes:

[0029] The first training module is used to fine-tune the msmarco-distilbert-cos-v5 model through the NSP task to obtain the correlation calculation model.

[0030] Optionally, in one embodiment of this application, it further includes:

[0031] The acquisition module is used to acquire the query information input by the user and the information of the most similar preset number of experts;

[0032] The second training module is used to obtain the similarity between the query information and the corresponding expert based on the query information and the expert's information. Based on the similarity, the GBrank model is trained.

[0033] To achieve the above objectives, a second aspect of the present invention proposes a retrieval method incorporating a knowledge graph, comprising:

[0034] The system retrieves query information input from the user; it segments the query information into words to obtain a list of words, expands these words using a knowledge graph to obtain an expanded list of words, calculates the weight of each word in the expanded list, sorts and filters the words based on their weights, and performs a search on the filtered words to obtain a preset number of expert information entries; it then sorts these preset number of experts and sends the information of the most relevant experts to the user based on the requested number of entries.

[0035] Send it to the user.

[0036] Optionally, in one embodiment of this application, the query information is segmented into words to obtain a list of words, and the words are expanded according to a knowledge graph to obtain an expanded list of words. The expanded list of words is then used to calculate...

[0037] Each word is assigned a weight, and the words are sorted and filtered based on their weights. A search is then performed on the filtered words to obtain information on a preset number of experts, including:

[0038] Use NLTK to segment the title and requirement text to obtain a list of words in the title and requirement text;

[0039] The vocabulary is expanded based on the word list and the subject knowledge graph, resulting in an expanded word list, and the source of each word in the list is marked.

[0040] The similarity between each word and the title and demand text is calculated using TF-IDF as the weight of each word; the weights of words sourced from the demand text are calculated and updated using the BERT model, and the weights of words sourced from the demand text are calculated using the DSSM model.

[0041] And update the weights of the words whose sources are labeled as titles;

[0042] The weights of words whose sources are labeled as knowledge graph extensions are reduced, and the weights are updated based on the reduced weights.

[0043] The words with updated weights are merged and sorted, and words with weights below a preset threshold are filtered and deleted.

[0044] Searching based on the filtered words yields a preset number of expert profiles.

[0045] 5. To achieve the above objectives, a third aspect of the present invention provides a computer device, including a memory and a processor.

[0046] and a computer program stored in memory and executable on a processor, wherein when the processor executes the computer program, it implements the knowledge graph retrieval method described in the above embodiments.

[0047] To achieve the above objectives, a fourth aspect of the present invention provides a non-transitory computer-readable storage medium.

[0048] When the instructions in the storage medium are executed by the processor, a retrieval method incorporating a knowledge graph can be performed.

[0049] This application provides embodiments of a knowledge graph-based retrieval system, method, computer device, and non-transitory computer-readable format.

[0050] The storage medium solves the technical problem that existing search engines struggle to accurately determine a user's true query intent and return query results. It uses Elasticsearch for storage and msmarco-distilbert-cos-v5 and GBrank technologies for sorting. Furthermore, when constructing a query, it incorporates a knowledge graph to build multi-path recall, achieving accurate retrieval and thus improving retrieval performance with strong scalability.

[0051] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0052] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0053] Figure 1 This is a schematic diagram of the structure of a retrieval system incorporating a knowledge graph, provided in Embodiment 1 of this application.

[0054] Figure 2 This is a flowchart of the recall module of the knowledge graph-based retrieval system according to an embodiment of this application;

[0055] Figure 3 This is a flowchart of the sorting module of the knowledge graph-based retrieval system according to an embodiment of this application;

[0056] Figure 4 This is a flowchart illustrating a retrieval method incorporating a knowledge graph, as provided in Embodiment 2 of this application. Detailed Implementation

[0057] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0058] Existing retrieval systems use an inverted index-based database structure, while this application uses Elasticsearch, a highly scalable approach. This application employs knowledge graphs for recommendations at the input level to expand the search query. Storage is handled by Elasticsearch, with sorting mechanisms created independently based on specific business requirements.

[0059] The following describes, with reference to the accompanying drawings, an embodiment of the knowledge graph-based retrieval system and method of this application.

[0060] Figure 1 This is a schematic diagram of the structure of a retrieval system incorporating a knowledge graph, provided in Embodiment 1 of this application.

[0061] like Figure 1 As shown, the retrieval system incorporating knowledge graphs includes:

[0062] Interaction module 10 is used to obtain query information input by the user.

[0063] Data storage module 20 is used to store expert information data using ElasticSearch;

[0064] The recall module 30 is used to segment the query information obtained by the interaction module to obtain a list of words, and expand the words according to the knowledge graph to obtain an expanded list of words. The weight of each word is calculated according to the expanded list of words, the words are sorted and filtered according to the weights, and the information of a preset number of experts is retrieved according to the filtered words.

[0065] The sorting module 40 is used to sort the preset number of experts obtained by the recall module, and send the information of the most relevant experts to the interaction module according to the number of user needs.

[0066] Interaction module 10 is also used to provide information from the most relevant experts to the user.

[0067] This application's embodiment of the knowledge graph-based retrieval system includes: an interaction module for acquiring query information input by the user; a data storage module for storing expert information data using Elasticsearch; a recall module for segmenting the query information acquired by the interaction module to obtain a list of words, expanding the words according to the knowledge graph to obtain an expanded list of words, calculating the weight of each word based on the expanded list, sorting and filtering the words according to the weights, and retrieving information from a preset number of experts based on the filtered words; a sorting module for sorting the preset number of experts obtained by the recall module, and sending the information of the most relevant experts to the interaction module according to the user's needs; and the interaction module for providing the information of the most relevant experts to the user. This solves the technical problem of existing search engines' difficulty in accurately determining the user's true query intent and returning query results. It uses Elasticsearch for storage, MSMarco-Distilbert-Cos-V5 and GBRank technologies for sorting, and incorporates a knowledge graph during query construction to build multi-path recall, achieving accurate retrieval, thereby improving retrieval performance and exhibiting strong scalability.

[0068] Input the given title, requirement description, and keywords, and then the backend logic will return the experts most relevant to the input information.

[0069] To efficiently solve expert retrieval on large datasets, this application comprehensively considers speed, recall, and ranking, and uses different methods for each, including:

[0070] Recall: When collecting expert information, use an Elasticsearch cluster to store data and provide fast retrieval.

[0071] Ranking: For the recalled experts, a series of features such as h-index, es relevance, and model relevance are used, and finally GBrank is used to fuse them to return the experts most relevant to the input title.

[0072] Speed: During the recall phase, Elasticsearch is used for storage, which allows for quick return of experts. During the sorting phase, which is more time-consuming, the experts' embeddings are saved in advance to improve speed.

[0073] Furthermore, in this embodiment, the data storage module is specifically used for:

[0074] For long fields in expert information data, ik_smart is used for word segmentation, and then the data is inserted into ElasticSearch for storage.

[0075] For short fields in expert information data, use keywords to insert them into Elasticsearch for storage.

[0076] The data storage module mainly designs the mapping configuration for inserting data. For fields such as experts' personal profiles, paper titles, educational backgrounds, and work backgrounds, ik_smart is used for word segmentation before insertion into Elasticsearch. For shorter phrases such as experts' research interests, the keyword method is used for insertion into Elasticsearch.

[0077] Furthermore, in this embodiment of the application, the query information includes a title and a request text, and the recall module is specifically used for:

[0078] Use NLTK to segment the title and requirement text to obtain a list of words in the title and requirement text;

[0079] The vocabulary is expanded based on the word list and the subject knowledge graph, resulting in an expanded word list, and the source of each word in the list is marked.

[0080] The similarity between each word and the title and demand text is calculated using TF-IDF as the weight of each word;

[0081] The weights of words sourced from the demand text are calculated and updated using the BERT model, and the weights of words sourced from the title are calculated and updated using the DSSM model.

[0082] The weights of words whose sources are labeled as knowledge graph extensions are reduced, and the weights are updated based on the reduced weights.

[0083] The words with updated weights are merged and sorted, and words with weights below a preset threshold are filtered and deleted.

[0084] Searching based on the filtered words yields a preset number of expert profiles.

[0085] This application recall module returns a preset number of experts from tens of millions of expert data points based on the input title and requirement description. The preset number can be 1000.

[0086] In this embodiment of the application, the weight of words whose source is marked as knowledge graph extension is reduced. For example, if a word comes from a knowledge graph, the weight of the word is reduced by 0.8 based on the current weight of the word.

[0087] like Figure 2As shown, the recall module includes: First, using NLTK to segment the input title and requirement description, the output is a list of words in the input content. The segmented words are then expanded based on the subject's knowledge graph, and each word is labeled with its source. TF-IDF is then used to calculate the similarity between each expanded word and the input content, serving as the weight for each word. The BERT model is then used to calculate and update the weights of words whose source is labeled as requirement text, and the DSSM model is used to calculate and update the weights of words whose source is labeled as title. The weighted words are then merged and sorted, and words with weights below a preset threshold are filtered and deleted. For words originating from the knowledge graph, only the top 10 are retained. The filtered and deleted words are then retrieved from the Interests database to return information from the top 1000 experts.

[0088] Furthermore, in this embodiment of the application, the query information also includes keywords and a sorting module, specifically used for:

[0089] The relevance of the title and a preset number of experts is calculated using a relevance calculation model.

[0090] Get the relevance of a preset number of experts returned by Elasticsearch;

[0091] By determining whether the keywords entered by the user match the keywords of a preset number of experts, the keyword matching value of the preset number of experts is obtained.

[0092] The model relevance, ElasticSearch relevance, keyword matching value, h-index, citation count, and paper publication count of a preset number of experts are input into the GBrank model for calculation to obtain the weights of the preset number of experts.

[0093] Based on keyword matching values, the weight of a preset number of experts is increased;

[0094] The experts are sorted according to their weighted values, and the information of the expert with the highest weight is sent to the interaction module according to the number of users' needs.

[0095] The sorting module in this application mainly sorts the 1,000 experts recalled by ElasticSearch and returns the experts most relevant to the input sentence according to the number of users required.

[0096] like Figure 3As shown, the ranking module mainly includes: calculating the relevance between the input title and the information of a preset number of experts obtained by the recall module using a model. The model used is msmarco-distilbert-cos-v5, which is then fine-tuned through the NSP task to learn the distribution of this task. Finally, both the input title and the expert's personal profile are represented as 512-dimensional vectors, and the similarity of the data is measured using these vectors. The model relevance, ElasticSearch relevance, keyword matching value, h-index, citation count, and paper publication count of the preset number of experts are input into the GBrank model for calculation to obtain the weights of the preset number of experts. The similarity between queries and experts is manually labeled. For each query, the 10 experts most similar to the query are returned. This data is used to train GBrank to measure the similarity between queries and experts.

[0097] Furthermore, in the embodiments of this application, it also includes:

[0098] The first training module is used to fine-tune the msmarco-distilbert-cos-v5 model through the NSP task to obtain the correlation calculation model.

[0099] Furthermore, in the embodiments of this application, it also includes:

[0100] The acquisition module is used to acquire the query information input by the user and the information of the most similar preset number of experts;

[0101] The second training module is used to obtain the similarity between the query information and the corresponding expert based on the query information and the expert's information, and to train the GBrank model based on the similarity.

[0102] Figure 4 This is a flowchart illustrating a retrieval method incorporating a knowledge graph, as provided in Embodiment 2 of this application.

[0103] like Figure 4 As shown, this knowledge graph-based retrieval method includes the following steps:

[0104] Step 401: Obtain the query information input by the user;

[0105] Step 402: Segment the query information into words to obtain a list of words, and expand the words according to the knowledge graph to obtain an expanded list of words. Calculate the weight of each word according to the expanded list of words, sort and filter the words according to the weights, and retrieve the information of a preset number of experts based on the filtered words.

[0106] Step 403: Sort the preset number of experts and send the information of the most relevant experts to the user terminal according to the number of users' needs.

[0107] This application's embodiment introduces a knowledge graph-based retrieval method. It obtains query information input from the user; segments the query information into words to obtain a list of words; expands these words using a knowledge graph to obtain an expanded list of words; calculates the weight of each word in the expanded list; sorts and filters the words based on their weights; and retrieves information from a preset number of experts based on the filtered words. The preset number of experts are then sorted, and the information of the most relevant experts is sent to the user based on the user's desired quantity. This solves the technical problem of existing search engines' difficulty in accurately determining the user's true query intent and returning query results. It uses Elasticsearch for storage, and MSMarco-Distilbert-Cos-V5 and GBRank technologies for sorting. Furthermore, by incorporating a knowledge graph during query construction and building multi-path recall, it achieves accurate retrieval, thereby improving retrieval performance and exhibiting strong scalability.

[0108] Furthermore, in this embodiment, the query information is segmented into words to obtain a list of words, and the words are expanded according to a knowledge graph to obtain an expanded list of words. The weight of each word is calculated based on the expanded list of words, and the words are sorted and filtered according to their weights. A search is then performed on the filtered words to obtain information on a preset number of experts, including:

[0109] Use NLTK to segment the title and requirement text to obtain a list of words in the title and requirement text;

[0110] The vocabulary is expanded based on the word list and the subject knowledge graph, resulting in an expanded word list, and the source of each word in the list is marked.

[0111] The similarity between each word and the title and demand text is calculated using TF-IDF as the weight of each word;

[0112] The weights of words sourced from the demand text are calculated and updated using the BERT model, and the weights of words sourced from the title are calculated and updated using the DSSM model.

[0113] The weights of words whose sources are labeled as knowledge graph extensions are reduced, and the weights are updated based on the reduced weights.

[0114] The words with updated weights are merged and sorted, and words with weights below a preset threshold are filtered and deleted.

[0115] Searching based on the filtered words yields a preset number of expert profiles.

[0116] To implement the above embodiments, the present invention also proposes a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the knowledge graph-based retrieval method described in the above embodiments.

[0117] To implement the above embodiments, the present invention also proposes a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the knowledge graph retrieval method of the above embodiments.

[0118] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0119] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0120] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0121] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0122] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0123] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0124] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0125] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A retrieval system incorporating knowledge graphs, characterized in that, include: The interaction module is used to obtain query information input by the user, including the title and the request text; The data storage module is used to store expert information data using Elasticsearch; The recall module is used to segment the query information obtained by the interaction module to obtain a list of words, and expand the words according to the knowledge graph to obtain an expanded list of words. The weight of each word is calculated according to the expanded list of words, the words are sorted and filtered according to the weights, and the filtered words are retrieved to obtain information on a preset number of experts. The sorting module is used to sort the preset number of experts obtained by the recall module, and send the information of the most relevant experts to the interaction module according to the number of user needs. The interaction module is also used to provide the information of the most relevant expert to the user terminal; The recall module is specifically used to segment the title and the demand text using NLTK to obtain a list of words in the title and the demand text. Based on the aforementioned word list and the subject's knowledge graph, expand the word list to obtain an expanded word list, and annotate the source of each word in the list; The similarity between each word and the title and requirement text is calculated using TF-IDF as the weight of each word; The weights of words sourced from the demand text are calculated and updated using the BERT model, and the weights of words sourced from the title are calculated and updated using the DSSM model. The weights of words whose sources are labeled as knowledge graph extensions are reduced, and the weights are updated based on the reduced weights. The words with updated weights are merged and sorted, and words with weights below a preset threshold are filtered and deleted. Searching based on the filtered words yields a preset number of expert profiles.

2. The system as described in claim 1, characterized in that, The data storage module is specifically used for: For long fields in expert information data, ik_smart is used for word segmentation, and then the data is inserted into ElasticSearch for storage. For short fields in expert information data, use keywords to insert them into Elasticsearch for storage.

3. The system as described in claim 1, characterized in that, The query information also includes keywords, and the sorting module is specifically used for: The relevance between the title and the preset number of experts is calculated using a relevance calculation model; Obtain the relevance of the preset number of experts returned by Elasticsearch; By determining whether the keywords input by the user match the keywords of the preset number of experts, the keyword matching value of the preset number of experts is obtained. The model relevance, ElasticSearch relevance, keyword matching value, h-index, citation count, and paper publication count of the preset number of experts are input into the GBrank model for calculation to obtain the weights of the preset number of experts. Based on the keyword matching value, the weight of the preset number of experts is increased; The preset number of experts are sorted according to their weighted values, and the information of the expert with the highest weight is sent to the interaction module according to the number of users' needs.

4. The system as described in claim 3, characterized in that, Also includes: The first training module is used to fine-tune the msmarco-distilbert-cos-v5 model through the NSP task to obtain the correlation calculation model.

5. The system as described in claim 3, characterized in that, Also includes: The acquisition module is used to acquire the query information input by the user and the information of the most similar preset number of experts; The second training module is used to obtain the similarity between the query information and the corresponding expert based on the query information and the expert information, and to train the GBrank model based on the similarity.

6. A retrieval method incorporating knowledge graphs, characterized in that, include: Obtain query information input by the user, including the title and the request text; The query information is segmented into words to obtain a list of words. The words are then expanded using a knowledge graph to obtain an expanded list of words. The weight of each word is calculated based on the expanded list of words. The words are sorted and filtered based on the weights. The filtered words are then retrieved to obtain information on a preset number of experts. This includes: using NLTK to segment the title and the request text to obtain a list of words for the title and the request text. Based on the aforementioned word list and the subject's knowledge graph, expand the word list to obtain an expanded word list, and annotate the source of each word in the list; The similarity between each word and the title and requirement text is calculated using TF-IDF as the weight of each word; The weights of words sourced from the demand text are calculated and updated using the BERT model, and the weights of words sourced from the title are calculated and updated using the DSSM model. The weights of words whose sources are labeled as knowledge graph extensions are reduced, and the weights are updated based on the reduced weights. The words with updated weights are merged and sorted, and words with weights below a preset threshold are filtered and deleted. Based on the filtered keywords, a search is conducted to obtain information on a predetermined number of experts; The preset number of experts are sorted, and the information of the most relevant experts is sent to the user terminal according to the number of users' needs.

7. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method as described in claim 6.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in claim 6.

Images