Academic literature retrieval method and device based on large model, storage medium and equipment

Abstract

The invention discloses an academic literature retrieval method and device based on a large model, a storage medium and computer equipment, and the method comprises the steps that academic literature retrieval information is obtained, and the academic literature retrieval information is a natural language text; a target retrieval formula corresponding to the academic literature retrieval information is generated through a retrieval scene large model, the retrieval scene large model is a large model for instruction fine tuning based on a fine tuning instruction set, and the fine tuning instruction set comprises a plurality of fine tuning instructions; each fine tuning instruction comprises an academic literature retrieval information sample and a corresponding retrieval type sample; and performing visual display on the target retrieval formula, and performing academic literature retrieval based on the target retrieval formula in response to a confirmation instruction for the target retrieval formula. According to the method and the device, the academic literature retrieval threshold is reduced, and the academic literature retrieval efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of document retrieval, and in particular to a method, device, storage medium and computer equipment for academic document retrieval based on a large model. Background Art

[0002] Traditional academic literature retrieval methods mainly rely on indexing or cataloging information. This method is limited to searching preset fields and it is difficult to go deep into the core content of the document. Despite the progress of information technology, traditional retrieval methods are still stretched in processing. They often stop at the basic attribute information of the document and it is difficult to touch the essence of the document content. What is more inadequate is that traditional methods are quite weak in supporting natural language semantic retrieval, which undoubtedly increases the requirements for the professional skills of search personnel and makes the document retrieval process more complicated and cumbersome. Summary of the invention

[0003] In view of this, the embodiments of the present application provide a large-model-based academic document retrieval method, apparatus, storage medium and computer equipment, which help to lower the threshold for document retrieval and improve document retrieval efficiency.

[0004] According to one aspect of the present application, a method for academic literature retrieval based on a large model is provided, the method comprising:

[0005] Obtaining academic document retrieval information, wherein the academic document retrieval information is a natural language text;

[0006] Generate a target search formula corresponding to the academic literature search information through a search scenario macromodel, wherein the search scenario macromodel is a macromodel that performs instruction fine-tuning based on a fine-tuning instruction set, wherein the fine-tuning instruction set includes a plurality of fine-tuning instructions, and each fine-tuning instruction includes an academic literature search information sample and a corresponding search formula sample;

[0007] The target search formula is displayed visually, and in response to a confirmation instruction for the target search formula, an academic literature search is performed based on the target search formula.

[0008] Optionally, obtaining academic literature retrieval information includes:

[0009] Access academic literature search information entered in text form; and / or,

[0010] Acquire academic document retrieval audio input in voice form, perform voice recognition on the academic document retrieval audio, and obtain academic document retrieval information in text form.

[0011] Optionally, the generating a target search formula corresponding to the academic literature search information by using the search scenario macro model includes:

[0012] Constructing a first prompt word based on the academic literature retrieval information, inputting the first prompt word into the retrieval scenario macromodel, so as to perform retrieval intent recognition based on the first prompt word through the retrieval scenario macromodel, and outputting retrieval element information;

[0013] A target search unit matching a preset search item is constructed based on the search element information, and the target search formula is generated based on the target search unit.

[0014] Optionally, constructing a first prompt word based on the academic literature retrieval information includes:

[0015] Based on the academic document retrieval information and a preset retrieval intent recognition template, construct the first prompt word, wherein the preset retrieval intent recognition template is used to guide the retrieval scenario macromodel to identify retrieval elements matching the preset retrieval item in the academic document retrieval information in the first prompt word, and to perform semantic understanding on the academic document retrieval information when the retrieval elements cannot be identified;

[0016] Accordingly, the step of constructing a target search unit matching a preset search item based on the search element information includes:

[0017] If the search element information is a search element corresponding to a preset search term, then based on the search elements in the search element information that match the preset search term, construct a target search unit corresponding to the preset search term;

[0018] If the search element information is semantic information corresponding to the academic literature search information, then based on the semantic information, a target search unit with a subject as a search item is constructed;

[0019] The preset search items include at least title, subject, publication year, author, and author unit.

[0020] Optionally, before generating a target search formula corresponding to the academic literature search information through the search scenario macro model, the method further includes:

[0021] Based on the academic document annotation items and bibliographic items, a document retrieval data item framework is constructed, wherein the document retrieval data item framework includes a plurality of retrieval units, each of which includes a preset retrieval item, a preset retrieval logic, and a preset retrieval item value rule;

[0022] Acquire multiple historical search logs, decompose each historical search log into search element samples matching the preset search terms, preset search logic, and preset search term value rules in the search unit based on the document search data item framework, and generate academic document search information samples corresponding to each historical search log;

[0023] Taking the search element sample as a search formula sample, constructing a fine-tuning instruction corresponding to each historical search log based on the academic document search information sample corresponding to each historical search log and the search formula sample, and generating a fine-tuning instruction set including a plurality of fine-tuning instructions;

[0024] The large model is fine-tuned based on the fine-tuning instruction set to obtain the retrieval scene large model.

[0025] Optionally, the visual display of the target search formula includes:

[0026] Based on the target search formula, fill in each search unit area and the search formula area in the search page, wherein the search unit area includes a search term, and search logic and search term value corresponding to the search term;

[0027] The populated search page is displayed.

[0028] Optionally, in response to a confirmation instruction for the target search formula, performing academic literature search based on the target search formula includes:

[0029] In response to a confirmation instruction on the search page, an academic literature search is performed based on the target search formula; or,

[0030] receiving a modification operation on the search page, modifying the filling content of the search unit area and the search formula area, and in response to a confirmation instruction on the modified search page, performing academic literature search based on the modified target search formula; or,

[0031] In response to a trigger operation on a search information entry control in the search page, the supplemented academic document search information is obtained, and a second prompt word is constructed based on the academic document search information and the supplemented academic document search system information, and the second prompt word is input into the search scenario big model to obtain a corrected target search formula, and the search page is filled with the corrected target search formula, and in response to a confirmation instruction on the search page or the modified search page, an academic document search is performed based on the confirmed target search formula.

[0032] According to another aspect of the present application, a large model-based academic document retrieval device is provided, the device comprising:

[0033] A search information acquisition module is used to acquire academic document search information, wherein the academic document search information is a natural language text;

[0034] A search formula generation module, used to generate a target search formula corresponding to the academic literature search information through a search scenario macromodel, wherein the search scenario macromodel is a macromodel for fine-tuning instructions based on a fine-tuning instruction set, wherein the fine-tuning instruction set includes a plurality of fine-tuning instructions, and each fine-tuning instruction includes an academic literature search information sample and a corresponding search formula sample;

[0035] The document retrieval module is used to visualize the target search formula and, in response to a confirmation instruction for the target search formula, perform academic document retrieval based on the target search formula.

[0036] Optionally, the retrieval information acquisition module is further used to:

[0037] Access academic literature search information entered in text form; and / or,

[0038] Acquire academic document retrieval audio input in voice form, perform voice recognition on the academic document retrieval audio, and obtain academic document retrieval information in text form.

[0039] Optionally, the search formula generating module is further used to:

[0040] Constructing a first prompt word based on the academic literature retrieval information, inputting the first prompt word into the retrieval scenario macromodel, so as to perform retrieval intent recognition based on the first prompt word through the retrieval scenario macromodel, and outputting retrieval element information;

[0041] A target search unit matching a preset search item is constructed based on the search element information, and the target search formula is generated based on the target search unit.

[0042] Optionally, the search formula generating module is further used to:

[0043] Based on the academic document retrieval information and a preset retrieval intent recognition template, construct the first prompt word, wherein the preset retrieval intent recognition template is used to guide the retrieval scenario macromodel to identify retrieval elements matching the preset retrieval item in the academic document retrieval information in the first prompt word, and to perform semantic understanding on the academic document retrieval information when the retrieval elements cannot be identified;

[0044] If the search element information is a search element corresponding to a preset search term, then based on the search elements in the search element information that match the preset search term, construct a target search unit corresponding to the preset search term;

[0045] If the search element information is semantic information corresponding to the academic literature search information, then based on the semantic information, a target search unit with a subject as a search item is constructed;

[0046] The preset search items include at least title, subject, publication year, author, and author unit.

[0047] Optionally, the device further comprises: a large model training module, configured to:

[0048] Based on the academic document annotation items and bibliographic items, a document retrieval data item framework is constructed, wherein the document retrieval data item framework includes a plurality of retrieval units, each of which includes a preset retrieval item, a preset retrieval logic, and a preset retrieval item value rule;

[0049] Acquire multiple historical search logs, decompose each historical search log into search element samples matching the preset search terms, preset search logic, and preset search term value rules in the search unit based on the document search data item framework, and generate academic document search information samples corresponding to each historical search log;

[0050] Taking the search element sample as a search formula sample, constructing a fine-tuning instruction corresponding to each historical search log based on the academic document search information sample corresponding to each historical search log and the search formula sample, and generating a fine-tuning instruction set including a plurality of fine-tuning instructions;

[0051] The large model is fine-tuned based on the fine-tuning instruction set to obtain the retrieval scene large model.

[0052] Optionally, the document retrieval module is further used to:

[0053] Based on the target search formula, fill in each search unit area and the search formula area in the search page, wherein the search unit area includes a search term, and search logic and search term value corresponding to the search term;

[0054] The populated search page is displayed.

[0055] Optionally, the document retrieval module is further used to:

[0056] In response to a confirmation instruction on the search page, an academic literature search is performed based on the target search formula; or,

[0057] receiving a modification operation on the search page, modifying the filling content of the search unit area and the search formula area, and in response to a confirmation instruction on the modified search page, performing academic literature search based on the modified target search formula; or,

[0058] In response to a trigger operation on a search information entry control in the search page, the supplemented academic document search information is obtained, and a second prompt word is constructed based on the academic document search information and the supplemented academic document search system information, and the second prompt word is input into the search scenario big model to obtain a corrected target search formula, and the search page is filled with the corrected target search formula, and in response to a confirmation instruction on the search page or the modified search page, an academic document search is performed based on the confirmed target search formula.

[0059] According to another aspect of the present application, a storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the above-mentioned academic document retrieval method based on a large model is implemented.

[0060] According to another aspect of the present application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor implements the above-mentioned large model-based academic document retrieval method when executing the program.

[0061] By means of the above technical solution, the embodiment of the present application provides an academic document retrieval method, device, storage medium and computer equipment based on a large model. Through fine-tuning of the instructions of the large model, the large model can more deeply understand and transform the user's natural language input, more accurately capture the user's intention, and generate a search formula that is more in line with user needs and closer to the content of the document, thereby improving the accuracy and depth of the search and solving the problem that traditional methods are difficult to reach the content of the document itself. Users do not need to have professional search skills or knowledge. This greatly reduces the threshold for search, allowing more users to easily perform academic document searches. In addition, by displaying the target search formula in a visual way, users can quickly confirm and search, which helps to improve the efficiency of the search and improve the accuracy of the search formula construction.

[0062] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0064] Figure 1 A schematic diagram of a process flow of a large model-based academic document retrieval method provided in an embodiment of the present application is shown;

[0065] Figure 2 A schematic diagram of a process flow of another large model-based academic document retrieval method provided in an embodiment of the present application is shown;

[0066] Figure 3 A schematic diagram of a document retrieval data item framework provided in an embodiment of the present application is shown;

[0067] Figure 4 A schematic diagram of a flow chart of a method for fine-tuning instructions for retrieving a large scene model provided in an embodiment of the present application is shown;

[0068] Figure 5 A schematic diagram of a process flow of another large model-based academic document retrieval method provided in an embodiment of the present application is shown;

[0069] Figure 6 A schematic diagram of the structure of an academic document retrieval device based on a large model provided in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0070] The present application will be described in detail below with reference to the accompanying drawings and in combination with embodiments. It should be noted that the embodiments and features in the embodiments of the present application can be combined with each other without conflict.

[0071] In this embodiment, a method for academic literature retrieval based on a large model is provided. Figure 1 As shown, the method includes:

[0072] Step 101: Obtain academic document retrieval information, wherein the academic document retrieval information is natural language text.

[0073] Step 102: Generate a target search formula corresponding to the academic literature retrieval information through a search scenario big model, wherein the search scenario big model is a big model that performs instruction fine-tuning based on a fine-tuning instruction set, and the fine-tuning instruction set includes multiple fine-tuning instructions, and each fine-tuning instruction includes an academic literature retrieval information sample and a corresponding search formula sample.

[0074] Step 103: Visually display the target search formula, and in response to a confirmation instruction for the target search formula, perform an academic literature search based on the target search formula.

[0075] The present application embodiment proposes an innovative academic document retrieval method, which makes full use of the ability of the large model to optimize the retrieval process. First, the user inputs a natural language text as academic document retrieval information, for example, the academic document retrieval information is "Please help me find the paper on single-atom catalysts by Academician B of University A". This academic document retrieval information only requires the searcher to enter it in the form of a simple natural language text, without explicitly stating which information is the search item, for example, there is no need for the searcher to enter a text similar to "Author is B, Author unit is A, Theme is single-atom catalyst", even an inexperienced searcher can easily provide retrieval information. Then, the natural language text entered by the user is processed by a fine-tuned retrieval scene large model. This large model has been fine-tuned by instructions, that is, a fine-tuning instruction set containing multiple fine-tuning instructions is used for training, and each fine-tuning instruction includes an academic document retrieval information sample and a corresponding optimized retrieval formula sample. In this way, the large model can learn how to convert the user's natural language input into a more accurate and efficient retrieval formula. Finally, the generated target retrieval formula will be displayed to the user in a visual way for the user to review and confirm. If the user confirms that it is correct, the system will perform an academic literature search based on this target search formula. If the user needs to make changes, he or she can make adjustments on the display interface.

[0076] By applying the technical solution of this embodiment and fine-tuning the instructions of the large model, the large model can more deeply understand and transform the user's natural language input, more accurately capture the user's intention, and generate a search formula that is more in line with user needs and closer to the document content, thereby improving the accuracy and depth of the search and solving the problem that traditional methods are difficult to reach the document content itself. Users do not need to have professional search skills or knowledge. This greatly reduces the threshold for search, allowing more users to easily search for academic documents. In addition, by displaying the target search formula in a visual way, users can quickly confirm and search, which helps to improve the efficiency of the search and improve the accuracy of the search formula construction.

[0077] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to fully illustrate the specific implementation process of this embodiment, another academic literature retrieval method based on a large model is provided, such as Figure 2 As shown, the method includes:

[0078] Step 201: construct a document retrieval data item framework based on academic document annotation items and bibliographic items, wherein the document retrieval data item framework includes multiple retrieval units, and each retrieval unit includes a preset retrieval item, a preset retrieval logic, and a preset retrieval item value rule.

[0079] In the embodiment of the present application, a document retrieval data item framework is constructed based on the annotation items (such as author, title, publication year, etc.) and bibliographic items (such as journal name, volume number, page number, etc.) of the academic document. This framework consists of multiple retrieval units, such as Figure 3 As shown, each search unit includes three key parts: preset search items (i.e., specific fields that users may want to search, such as "author name"), preset search logic (such as "contains", "equal to", "starts with", "exact match", "fuzzy match", etc., used to define the relationship between search items and search values), and preset search item value rules (such as values ​​or value ranges, etc., used to specify the input requirements and matching methods of search item values). The embodiments of the present application make the search process more structured by defining clear search units and rules, which helps to improve the accuracy and efficiency of retrieval.

[0080] Step 202, obtain multiple historical search logs, decompose each historical search log into a search element sample that matches the preset search item, preset search logic, and preset search item value rules in the search unit based on the document search data item framework, and generate an academic document search information sample corresponding to each historical search log.

[0081] Step 203: Use the retrieval element samples as retrieval formula samples, construct fine-tuning instructions corresponding to each historical retrieval log based on the academic document retrieval information samples corresponding to each historical retrieval log and the retrieval formula samples, and generate a fine-tuning instruction set including multiple fine-tuning instructions; perform instruction fine-tuning on the large model based on the fine-tuning instruction set to obtain the retrieval scenario large model.

[0082] In the present application embodiment, Figure 4As shown in FIG. 1 , multiple historical search logs (log data) are obtained, which record the user's past search behavior and results. Then, based on the literature search data item framework, each historical search log is decomposed into search element samples (such as specific author names, title keywords, etc.) that match the search unit, and the corresponding academic literature search information sample is generated. This academic literature search information sample is a natural language text containing each search element sample, and the generated search element sample is used as a search formula sample. Combined with the academic literature search information sample, a fine-tuning instruction is constructed. For example, the search element sample is: the author is Li Yadong, the author's unit is Tsinghua University, and the subject is the preparation method of single-atom catalysts; the generated academic literature search information sample can be: Please help me find the paper on single-atom catalysts by Academician Li Yadong of Tsinghua University. The academic literature search information sample can be manually written based on the historical search log or the search element sample, or it can be automatically generated by the large model and manually confirmed. Each fine-tuning instruction contains a specific search scenario (i.e., an academic literature search information sample) and the expected output (i.e., the corresponding search formula sample, generally output as search element information). By collecting multiple such fine-tuning instructions, a fine-tuning instruction set is formed. Then, based on this instruction set, the large model is fine-tuned to obtain a large model of retrieval scenarios specifically for academic literature retrieval. The embodiment of the present application can obtain the user's real retrieval behavior and needs through the analysis of historical retrieval logs, providing valuable data support for subsequent model fine-tuning. The generated retrieval factor samples and academic literature retrieval information samples cover a variety of retrieval scenarios and needs, which helps to improve the generalization ability of the model. Through the training of the fine-tuning instruction set, the large model can learn how to convert the user's natural language input into a structured retrieval formula to achieve accurate recognition of retrieval intent.

[0083] Step 204: obtain academic document retrieval information input in text form; and / or obtain academic document retrieval audio input in voice form, and perform voice recognition on the academic document retrieval audio to obtain academic document retrieval information in text form.

[0084] In the embodiment of the present application, academic literature search information input by the user in text form or voice form can be received. If it is voice input, the system will first perform voice recognition and convert it into search information in text form. The embodiment of the present application supports both text and voice input methods, which meets the input habits and needs of different users. The application of voice recognition technology allows users to search through voice commands, improving the convenience of operation and user experience.

[0085] Step 205: construct a first prompt word based on the academic literature retrieval information, input the first prompt word into the retrieval scenario model, identify the retrieval intent based on the first prompt word through the retrieval scenario model, and output retrieval element information; construct a target retrieval unit that matches the preset retrieval item based on the retrieval element information, and generate the target retrieval formula based on the target retrieval unit.

[0086] In an embodiment of the present application, a first prompt word is constructed based on the academic literature retrieval information input by the user, and is input into the retrieval scenario large model. The retrieval scenario large model will perform retrieval intent recognition based on the first prompt word, and output retrieval element information (such as the author, title, etc. that the user wants to retrieve). Then, based on these retrieval element information, a target retrieval unit that matches the preset retrieval item is constructed, and a target retrieval formula is generated. Through the retrieval intent recognition capability of the large model, the embodiment of the present application can accurately capture the user's retrieval needs and provide accurate direction for subsequent retrieval operations. The intelligent generation of the target retrieval formula reduces the cumbersome process of the user manually constructing the retrieval formula and improves the retrieval efficiency.

[0087] In an embodiment of the present application, optionally, constructing the first prompt word based on the academic document retrieval information includes: constructing the first prompt word based on the academic document retrieval information and a preset retrieval intent recognition template, wherein the preset retrieval intent recognition template is used to guide the retrieval scenario macro model to identify a retrieval element matching the preset retrieval item in the academic document retrieval information in the first prompt word, and to perform semantic understanding on the academic document retrieval information when the retrieval element cannot be identified;

[0088] Correspondingly, the target search unit matching the preset search item based on the search element information is constructed, including: if the search element information is a search element corresponding to the preset search item, then based on the search elements in the search element information that match the preset search item, a target search unit corresponding to the preset search item is constructed; if the search element information is semantic information corresponding to the academic literature retrieval information, then based on the semantic information, a target search unit with the subject as the search item is constructed; wherein the preset search item includes at least title, subject, publication year, author, and author unit.

[0089] In this embodiment, when constructing the first prompt word, the system not only relies on the academic literature retrieval information input by the user, but also combines the preset retrieval intent recognition template. This template is a structured framework used to guide the retrieval scenario large model to accurately identify the retrieval elements that match the preset retrieval items (such as title, subject, publication year, author, author unit, etc.) in the first prompt word. These retrieval elements are explicitly specified by the user, or can be inferred from the user input through the guidance of the template. If the retrieval scenario large model cannot directly identify the retrieval elements that match the preset retrieval items in the first prompt word (for example, the user inputs a natural language description instead of a clear keyword or phrase), the template will guide the retrieval scenario large model to trigger the semantic understanding mechanism. This mechanism allows the large model to perform a deeper semantic analysis of the user input to extract key information related to the retrieval intent. The preset retrieval intent recognition template in the embodiment of the present application provides clear guidance for the retrieval scenario large model, which helps the model to more accurately identify the user's retrieval intent. In the case where the retrieval elements cannot be directly identified, the semantic understanding mechanism allows the model to perform a deeper analysis of the user input, thereby improving the flexibility and accuracy of the retrieval. Further, if Figure 5 As shown in the figure, when constructing a target search unit based on the search element information, the system will adopt different strategies according to the nature of the search element information. If the search element information directly corresponds to the preset search item (for example, the user explicitly specifies the author's name or the article title), the target search unit corresponding to the preset search item will be constructed based on these search elements. These target search units will be directly used for subsequent search operations. If the search element corresponding to the preset search item is not identified from the academic literature search information, that is, the search element information is extracted through the semantic understanding mechanism (that is, they are not directly specified by the user, but obtained by analyzing the user input), then as a bottoming strategy, a target search unit with the subject as the search item will be constructed. This is because the semantic understanding mechanism can usually capture the core topic or theme in the user input, and the subject search item can flexibly match the academic literature containing these topics or themes. For example, if the academic literature search information is "single-atom catalyst is a hot topic", then the information can be semantically understood through the search scene big model, and "single-atom catalyst" can be output, thereby constructing a target search unit with the search theme of single-atom catalyst. The embodiments of the present application can flexibly construct a target retrieval unit according to the nature of the retrieval element information, and support both direct matching of user-specified retrieval elements and subject information extracted through a semantic understanding mechanism. By accurately matching preset retrieval items or subject information extracted using a semantic understanding mechanism, the system can construct a more accurate target retrieval unit, thereby improving the accuracy of the retrieval.

[0090] Step 206: based on the target search formula, fill in each search unit area and search formula area in the search page, wherein the search unit area includes the search terms, and the search logic and search term value corresponding to the search terms; and display the filled search page.

[0091] In an embodiment of the present application, after determining the target search formula, the various search unit areas and search formula areas in the search page can be filled in according to the target search formula. These areas include search terms (such as "author", "title", etc.), search logic (such as "contains", "equal to", etc.) and search term values ​​(i.e., the specific content that the user wants to retrieve). After filling is completed, the search page is displayed to the user for viewing. Through the visual search page, the user can intuitively see the selection and configuration of the search terms, which helps the user to confirm and adjust the search requirements. The user can manually adjust the search terms and search logic on the search page to achieve more accurate search operations.

[0092] Step 207: In response to a confirmation instruction for the target search formula, an academic literature search is performed based on the target search formula.

[0093] In the embodiment of the present application, if the user confirms that the target search formula is correct (which can be confirmed by viewing the search page in step 206), academic literature search can be performed based on the target search formula. The embodiment of the present application performs search based on the structured target search formula, which can significantly improve the accuracy of the search and reduce the interference of irrelevant results.

[0094] In a specific application scenario, the system receives user input information: Please help me find the paper on single-atom catalysts by Academician Li Yadong of Tsinghua University. The search scenario model analyzes the user input, understands the search intent for the paper, and extracts the key information elements input by the user: Tsinghua University, Li Yadong, and single-atom catalyst. Combined with the search intent, the key information elements are used to automatically construct three search units: author, author unit, and subject. According to the logical relationship of the key information elements in the search intent, they are automatically assembled into search expressions and visualized. Users can modify the search unit or re-enter the search statement according to their needs. After re-entering the content, the search scenario model re-executes the intention analysis and key information element extraction work. For example, the user inputs: modify the query subject to single-atom catalyst preparation method. The search domain model recognizes the user's intention to modify the subject search unit, extracts the key information element: single-atom catalyst preparation method, and automatically modifies the subject search unit content to "single-atom catalyst preparation method". After the user confirms, "Author = Li Yadong and Author Unit = Tsinghua University and Subject = Single-atom Catalyst Preparation Method" is used as the search condition to perform the search operation in the academic resource database.

[0095] In an embodiment of the present application, optionally, step 207 includes: in response to a confirmation instruction on the search page, performing an academic literature search based on the target search formula; or, receiving a modification operation on the search page, modifying the fill content of the search unit area and the search formula area, and in response to a confirmation instruction on the modified search page, performing an academic literature search based on the modified target search formula; or, in response to a trigger operation on a search information entry control in the search page, obtaining supplemented academic literature search information, and constructing a second prompt word based on the academic literature search information and the supplemented academic literature search system information, inputting the second prompt word into the search scenario big model to obtain a corrected target search formula, filling the search page with the corrected target search formula, and in response to a confirmation instruction on the search page or the modified search page, performing an academic literature search based on the confirmed target search formula.

[0096] In this embodiment, in the process of carrying out academic document retrieval based on target search formula, different responses are given according to the operation of the search page by the user. When the user confirms that the content on the search page is correct and issues a confirmation instruction, academic document retrieval is directly carried out based on the current target search formula. That is, the target search formula in the search formula area on the search page is used to construct a search request, and sent to the academic document database for retrieval. If the user finds that the content on the search page is wrong or needs to be adjusted, the filling content of the search unit area and the search formula area can be changed by modifying the operation. After the modification is completed, when the user confirms the modified search page again, academic document retrieval is carried out based on this updated target search formula. In addition, the search page can also include a search information input control, such as a text box, a drop-down menu, etc., allowing the user to enter or select additional academic document retrieval information. After the user triggers these controls and enters the information, the system can construct a second prompt word based on the information of the original academic document retrieval information and the user's supplementary recording. This second prompt word is subsequently input into the search scene macromodel to obtain a corrected target search formula. The corrected target search formula will be filled in the search page, waiting for the user's final confirmation. Once the user confirms, the system will perform academic literature retrieval based on this confirmed target search formula. It should be noted that the embodiment of the present application supports multiple rounds of interactive corrections, until the user completes the correction and clicks the search button, and the target search formula finally confirmed is submitted to the database for retrieval operation. The embodiment of the present application provides a variety of user interaction methods, allowing the user to directly confirm the search page, modify the search page or enter additional search information. This interactivity enables the user to flexibly adjust the search request according to their needs, and by allowing the user to modify and supplement the search information, it can be ensured that the target search formula ultimately used for retrieval is more accurate and meets the user's expectations. In addition, the system can handle different forms of user input and confirmation instructions, and provides a flexible search process to adapt to the search habits and needs of different users.

[0097] By applying the technical solution of this embodiment, through the automated and intelligent retrieval process, the time for users to manually input and screen documents is reduced, and the retrieval efficiency is significantly improved. The preset retrieval intention recognition template and semantic understanding mechanism enable the system to understand the user's retrieval needs more quickly and construct a retrieval formula that meets the user's expectations. It is possible to construct an accurate target retrieval unit based on the academic document retrieval information and preset retrieval items input by the user, thereby improving the accuracy of the retrieval and further ensuring the accuracy of the final retrieval results. By providing a variety of user interaction methods, such as directly confirming the retrieval page, modifying the retrieval page, entering additional retrieval information, etc., by allowing users to modify and supplement retrieval information, and using the retrieval scenario large model for correction, users can flexibly adjust retrieval requests according to their needs. The responsive interface design and intuitive operation process reduce the user's learning cost and improve the user experience.

[0098] Further, as Figure 1 The specific implementation of the method, the embodiment of the present application provides an academic document retrieval device based on a large model, such as Figure 6 As shown, the device comprises:

[0099] A search information acquisition module is used to acquire academic document search information, wherein the academic document search information is a natural language text;

[0100] A search formula generation module, used to generate a target search formula corresponding to the academic literature search information through a search scenario macromodel, wherein the search scenario macromodel is a macromodel for fine-tuning instructions based on a fine-tuning instruction set, wherein the fine-tuning instruction set includes a plurality of fine-tuning instructions, and each fine-tuning instruction includes an academic literature search information sample and a corresponding search formula sample;

[0101] The document retrieval module is used to visualize the target search formula and, in response to a confirmation instruction for the target search formula, perform academic document retrieval based on the target search formula.

[0102] Optionally, the retrieval information acquisition module is further used to:

[0103] Access academic literature search information entered in text form; and / or,

[0104] Acquire academic document retrieval audio input in voice form, perform voice recognition on the academic document retrieval audio, and obtain academic document retrieval information in text form.

[0105] Optionally, the search formula generating module is further used to:

[0106] Constructing a first prompt word based on the academic literature retrieval information, inputting the first prompt word into the retrieval scenario macromodel, so as to perform retrieval intent recognition based on the first prompt word through the retrieval scenario macromodel, and outputting retrieval element information;

[0107] A target search unit matching a preset search item is constructed based on the search element information, and the target search formula is generated based on the target search unit.

[0108] Optionally, the search formula generating module is further used to:

[0109] Based on the academic document retrieval information and a preset retrieval intent recognition template, construct the first prompt word, wherein the preset retrieval intent recognition template is used to guide the retrieval scenario macromodel to identify retrieval elements matching the preset retrieval item in the academic document retrieval information in the first prompt word, and to perform semantic understanding on the academic document retrieval information when the retrieval elements cannot be identified;

[0110] If the search element information is a search element corresponding to a preset search term, then based on the search elements in the search element information that match the preset search term, construct a target search unit corresponding to the preset search term;

[0111] If the search element information is semantic information corresponding to the academic literature search information, then based on the semantic information, a target search unit with a subject as a search item is constructed;

[0112] The preset search items include at least title, subject, publication year, author, and author unit.

[0113] Optionally, the device further comprises: a large model training module, configured to:

[0114] Based on the academic document annotation items and bibliographic items, a document retrieval data item framework is constructed, wherein the document retrieval data item framework includes a plurality of retrieval units, each of which includes a preset retrieval item, a preset retrieval logic, and a preset retrieval item value rule;

[0115] Acquire multiple historical search logs, decompose each historical search log into search element samples matching the preset search terms, preset search logic, and preset search term value rules in the search unit based on the document search data item framework, and generate academic document search information samples corresponding to each historical search log;

[0116] Taking the search element sample as a search formula sample, constructing a fine-tuning instruction corresponding to each historical search log based on the academic document search information sample corresponding to each historical search log and the search formula sample, and generating a fine-tuning instruction set including a plurality of fine-tuning instructions;

[0117] The large model is fine-tuned based on the fine-tuning instruction set to obtain the retrieval scene large model.

[0118] Optionally, the document retrieval module is further used to:

[0119] Based on the target search formula, fill in each search unit area and the search formula area in the search page, wherein the search unit area includes a search term, and search logic and search term value corresponding to the search term;

[0120] The populated search page is displayed.

[0121] Optionally, the document retrieval module is further used to:

[0122] In response to a confirmation instruction on the search page, an academic literature search is performed based on the target search formula; or,

[0123] receiving a modification operation on the search page, modifying the filling content of the search unit area and the search formula area, and in response to a confirmation instruction on the modified search page, performing academic literature search based on the modified target search formula; or,

[0124] In response to a trigger operation on a search information entry control in the search page, the supplemented academic document search information is obtained, and a second prompt word is constructed based on the academic document search information and the supplemented academic document search system information, and the second prompt word is input into the search scenario big model to obtain a corrected target search formula, and the search page is filled with the corrected target search formula, and in response to a confirmation instruction on the search page or the modified search page, an academic document search is performed based on the confirmed target search formula.

[0125] It should be noted that for other corresponding descriptions of the functional units involved in the large model-based academic document retrieval device provided in the embodiment of the present application, please refer to Figures 1 to 5 The corresponding description in the method will not be repeated here.

[0126] The embodiment of the present application also provides a computer device, which can be a personal computer, a server, a network device, etc. The computer device includes a bus, a processor, a memory and a communication interface, and can also include an input and output interface and a display device. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store location information. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the steps in each method embodiment are implemented.

[0127] Those skilled in the art will appreciate that the structure of the above-mentioned computer device is only a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components, or combine certain components, or have a different arrangement of components.

[0128] In one embodiment, a computer-readable storage medium is provided. The computer-readable storage medium may be non-volatile or volatile, and stores a computer program thereon. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0129] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0130] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0131] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0132] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0133] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for academic literature retrieval based on a large model, characterized in that: The method comprises: Obtaining academic document retrieval information, wherein the academic document retrieval information is a natural language text; Generate a target search formula corresponding to the academic literature search information through a search scenario macromodel, wherein the search scenario macromodel is a macromodel that performs instruction fine-tuning based on a fine-tuning instruction set, wherein the fine-tuning instruction set includes a plurality of fine-tuning instructions, and each fine-tuning instruction includes an academic literature search information sample and a corresponding search formula sample; The target search formula is displayed visually, and in response to a confirmation instruction for the target search formula, an academic literature search is performed based on the target search formula.

2. The method according to claim 1, characterized in that The obtaining of academic literature search information includes: Access academic literature search information entered in text form; and / or, Acquire academic document retrieval audio input in voice form, perform voice recognition on the academic document retrieval audio, and obtain academic document retrieval information in text form.

3. The method according to claim 1, characterized in that The target search formula corresponding to the academic literature search information is generated by using the search scenario macro model, including: Constructing a first prompt word based on the academic literature retrieval information, inputting the first prompt word into the retrieval scenario macromodel, so as to perform retrieval intent recognition based on the first prompt word through the retrieval scenario macromodel, and outputting retrieval element information; A target search unit matching a preset search item is constructed based on the search element information, and the target search formula is generated based on the target search unit.

4. The method according to claim 3, characterized in that The step of constructing a first prompt word based on the academic literature retrieval information includes: Based on the academic document retrieval information and a preset retrieval intent recognition template, construct the first prompt word, wherein the preset retrieval intent recognition template is used to guide the retrieval scenario macromodel to identify retrieval elements matching the preset retrieval item in the academic document retrieval information in the first prompt word, and to perform semantic understanding on the academic document retrieval information when the retrieval elements cannot be identified; Accordingly, the step of constructing a target search unit matching a preset search item based on the search element information includes: If the search element information is a search element corresponding to a preset search term, then based on the search elements in the search element information that match the preset search term, construct a target search unit corresponding to the preset search term; If the search element information is semantic information corresponding to the academic literature search information, then based on the semantic information, a target search unit with a subject as a search item is constructed; The preset search items include at least title, subject, publication year, author, and author unit.

5. The method according to claim 4, characterized in that Before generating the target search formula corresponding to the academic literature search information through the search scenario macro model, the method further includes: Based on the academic document annotation items and bibliographic items, a document retrieval data item framework is constructed, wherein the document retrieval data item framework includes a plurality of retrieval units, each of which includes a preset retrieval item, a preset retrieval logic, and a preset retrieval item value rule; Acquire multiple historical search logs, decompose each historical search log into search element samples matching the preset search terms, preset search logic, and preset search term value rules in the search unit based on the document search data item framework, and generate academic document search information samples corresponding to each historical search log; Taking the search element sample as a search formula sample, constructing a fine-tuning instruction corresponding to each historical search log based on the academic document search information sample corresponding to each historical search log and the search formula sample, and generating a fine-tuning instruction set including a plurality of fine-tuning instructions; The large model is fine-tuned based on the fine-tuning instruction set to obtain the retrieval scene large model.

6. The method according to claim 1, characterized in that The visual display of the target search formula includes: Based on the target search formula, fill in each search unit area and the search formula area in the search page, wherein the search unit area includes a search term, and search logic and search term value corresponding to the search term; The populated search page is displayed.

7. The method according to claim 6, characterized in that The step of performing academic literature search based on the target search formula in response to a confirmation instruction of the target search formula includes: In response to a confirmation instruction on the search page, an academic literature search is performed based on the target search formula; or, receiving a modification operation on the search page, modifying the filling content of the search unit area and the search formula area, and in response to a confirmation instruction on the modified search page, performing academic literature search based on the modified target search formula; or, In response to a trigger operation on a search information entry control in the search page, the supplemented academic document search information is obtained, and a second prompt word is constructed based on the academic document search information and the supplemented academic document search system information, and the second prompt word is input into the search scenario big model to obtain a corrected target search formula, and the search page is filled with the corrected target search formula, and in response to a confirmation instruction on the search page or the modified search page, an academic document search is performed based on the confirmed target search formula.

8. An academic document retrieval device based on a large model, characterized in that: The device comprises: A search information acquisition module is used to acquire academic document search information, wherein the academic document search information is a natural language text; A search formula generation module, used to generate a target search formula corresponding to the academic literature search information through a search scenario macromodel, wherein the search scenario macromodel is a macromodel for fine-tuning instructions based on a fine-tuning instruction set, wherein the fine-tuning instruction set includes a plurality of fine-tuning instructions, and each fine-tuning instruction includes an academic literature search information sample and a corresponding search formula sample; The document retrieval module is used to visualize the target search formula and, in response to a confirmation instruction for the target search formula, perform academic document retrieval based on the target search formula.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device comprising a storage medium, a processor, and a computer program stored in the storage medium and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images