Book recommendation method and system based on vertical large model

Abstract

The invention discloses a book recommendation method and system based on a vertical large model, and the system comprises a user feature extraction module, a book feature extraction module, a user real-time preference extraction module, a vector library storage module, a label matching module, a book recommendation reading understanding module, and a user question module. The invention aims to provide personalized book recommendation service for users. A new book recommendation method is provided on the basis of a self-developed data training vertical large model, the self-developed vertical large model has rich book knowledge reserve and natural language understanding ability, real demands of users can be accurately informed, personalized books are recommended to the users, and intelligent and accurate book recommendation is truly achieved.

Description

Technical Field

[0001] The invention relates to the technical field of computer science and is applied to a book recommendation system. Background Art

[0002] In the field of book recommendation, the core goal is to provide users with accurate and personalized book recommendation services. Traditional recommendation methods mainly rely on analyzing users' historical behavior data, such as the list of books they have read and the reviews they have left, etc., and use recommendation algorithms to predict users' potential preferences and match them with books of different styles in the massive library, so as to push books that users may be interested in and meet their personalized needs. However, traditional methods have obvious limitations. They can only make recommendations based on established tags and lack the ability to make real-time dynamic recommendations based on users' natural language instructions and to correct the recommendation direction in a timely manner. This limits the accuracy and flexibility of recommendations to a certain extent.

[0003] In order to effectively overcome this difficult problem in the field of book recommendations, the present invention innovatively integrates cutting-edge big model technology, conducts training based on proprietary data, and creates a professional vertical big model. Relying on the powerful capabilities of this self-developed vertical big model, the present invention has carefully designed an advanced book recommendation system and proposed a new book recommendation method. This self-developed vertical big model has a rich reserve of book knowledge and natural language understanding capabilities, can accurately understand the real needs of users, and recommend highly personalized book lists to users, truly realizing intelligent and precise book recommendations, and opening a new chapter in personalized book recommendations. Summary of the invention

[0004] Aiming at the shortcomings of the existing detection methods, a book recommendation system based on a vertical large model is proposed. The system includes: user feature extraction module, book feature extraction module, user real-time preference extraction module, vector library storage module, label matching module, book recommendation reading comprehension module and user questioning module.

[0005] The user feature extraction module accurately extracts key features from massive user behavior data, mines multi-dimensional information such as reading preferences, comment sentiment tendencies and focus points, search history exploration intentions, and converts them into concise and accurate user potential preference labels.

[0006] The book feature extraction module deeply analyzes and labels the multi-source information of the book, identifies key elements such as the core content of the abstract, the subject knowledge field, the style and emotional color and artistic characteristics, the author's creative style, and converts them into clear and accurate book labels.

[0007] The user real-time preference extraction module captures the user's current needs and preferences in real time, combines potential preference tags, keenly perceives changes in needs under specific circumstances, and accurately extracts real-time preference tags, so that the recommendation system can adjust the recommendation direction in a timely manner.

[0008] The vector library storage module efficiently stores and manages book feature vectors, provides structured and easily searchable storage space for book tags and related information, builds a huge book feature vector library, and supports fast vector search and matching operations.

[0009] The tag matching module uses an efficient vector matching algorithm based on the user's real-time preference tags to quickly retrieve book tags with high matching degrees, recall books that meet the user's needs, extract detailed information, and provide an accurate set of book candidates for generating personalized recommended book lists.

[0010] The book recommendation reading comprehension module combines the user's real-time preference tags and recalled book information, analyzes the points of convergence through deep reading comprehension and text generation capabilities, explores the highlights of book value, and generates attractive recommended book lists and recommendation words that meet personalized needs.

[0011] The user inquiry module processes user inquiries and optimizes recommendation results, integrates additional requirements with original questions, reactivates related modules, regenerates recommended book lists and recommendation statements according to optimization requirements, and supports iterative inquiry and recommendation until the user is satisfied.

[0012] In addition, the present invention also relates to a book recommendation method based on a vertical large model, which is applied to the book recommendation system of the vertical large model, and comprises the following steps: Step S1: Based on the user's book list, user comments, and user search history information, a vertical large model is used to extract user features and generate user potential preference tags; Step S2: Based on the book summary, book theme, book style, and book author information, the vertical large model is used to extract book features, generate book tags, and store relevant information in a vector library for subsequent book display; Step S3: Combine the user's real questions and the user's potential preference tags, use the vertical large model to extract the user's real-time preference tags; Step S4: Match the user's actual preference label with the book label in the vector library, recall the books with a high matching degree, and return the relevant information of the book from the vector library; Step S5: Combine the user's real-time preference tags in step S3 and the book-related information recalled in step S4, use the vertical large model to perform reading comprehension, and generate a personalized recommended book list and personalized recommendation words; Step S6: If the user has further requirements for the personalized recommended book list generated in step S5, the user can express the new requirements by asking questions, combining the user's additional requirements with the user's original questions, regenerating accurate real user questions, and re-going through the process of step S3-step S5 to regenerate the personalized recommended book list and personalized recommendation words based on the optimized user requirements. This step supports unlimited iterations of asking questions and recommendations. The system will continue to adjust the recommendation strategy based on user feedback until the user is completely satisfied with the recommendation results, which fully reflects the flexibility and user orientation of the recommendation system, ensuring that the recommendation service can continue to fit the user's changing personalized needs and provide users with the ultimate personalized book recommendation experience. Beneficial Effects

[0013] The present invention makes full use of the powerful capabilities of the vertical large model and exhibits many significant advantages: 1. Deeply understand users and books Accurately grasp user needs: With the training of massive book data, the vertical large model can deeply understand the fuzzy needs input by users and accurately locate their specific preferences. For example, when a user mentions "I like historical books", the model not only recognizes the keyword "history", but also combines the user's past behavior to accurately determine whether he prefers historical figures, event interpretations, historical novels and other sub-fields, laying the foundation for personalized recommendations.

[0014] Comprehensive analysis of book features: When extracting book features, the vertical large model can accurately identify multi-source information such as book summaries, themes, styles, and author styles, and convert complex book content into clear labels. This enables the system to carefully distinguish between books on different historical themes, such as accurately classifying historical books that focus on war descriptions and those that focus on cultural heritage, providing strong support for accurate matching.

[0015] 2. Enhance the flexibility and interactivity of the recommendation system Real-time preference capture and adjustment: The user real-time preference extraction module uses real-time semantic analysis of vertical large models to closely focus on the user's real problems, keenly perceive changes in demand, and adjust the recommendation direction in a timely manner. When a user asks "What science fiction novels are suitable for teenagers recently", the system quickly captures real-time preferences such as "teenagers" and "science fiction novels", and accurately recommends suitable books to meet the user's immediate needs.

[0016] Follow-up questions to optimize recommendation results: The user follow-up question module supports unlimited iterations of follow-up questions, integrating the user's additional requirements with the original questions, reactivating relevant modules, and generating new recommended book lists and recommendation words based on optimization requirements. This flexibility and user orientation ensures that the recommendation service continues to meet the changing needs of users until the users are satisfied, providing users with an extremely personalized experience.

[0017] In summary, this method uses the deep learning, semantic understanding and real-time analysis capabilities of the vertical large model to achieve accurate grasp of user needs, comprehensive analysis of book features and continuous optimization of recommendation results, bringing users accurate, real-time and personalized book recommendation experience, and promoting the book recommendation system to a higher level of development. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 The present invention is a flowchart of a book recommendation system based on a vertical large model. DETAILED DESCRIPTION

[0019] Embodiment 1 of the present invention relates to a book recommendation system based on a vertical large model, as shown in the attached Figure 1 As shown, the system includes: a user feature extraction module, a book feature extraction module, a user real-time preference extraction module, a vector library storage module, a label matching module, a book recommendation reading comprehension module, and a user questioning module.

[0020] The user feature extraction module has the function of accurately extracting key features from massive user behavior data. It can deeply mine multi-dimensional information such as the reading preferences reflected in the user's book list, the emotional tendencies and concerns contained in the user's comments, and the potential exploration intentions behind the user's search history. With the deep learning ability of the vertical large model, it can transform these complex information into concise, accurate and representative user potential preference labels, providing accurate user portrait basis for personalized recommendations.

[0021] The vertical large model is trained based on massive data, and can deeply understand the fuzzy needs of users' input and accurately locate their specific preferences. The vertical large model has deep language understanding capabilities and can support multiple tasks. It is used in the user feature extraction module, book feature extraction module, user real-time preference extraction module, book recommendation reading comprehension module, and user questioning module. According to the different inputs and instructions of each module, it generates results that comply with the instruction requirements, thereby serving multiple modules and providing strong technical support and core power for realizing personalized and precise book recommendations.

[0022] The book feature extraction module has the function of deeply analyzing and labeling multi-source information of books. It can accurately identify key elements such as the core content in the book summary, the knowledge field covered by the book theme, the emotional color and artistic characteristics reflected in the book style, and the unique creative style of the book author, and convert these elements into clear and accurate book labels through the semantic understanding and feature extraction technology of the vertical large model.

[0023] The user real-time preference extraction module has the function of capturing the user's current needs and preferences in real time. It can closely focus on the real questions raised by the user, combine the user's existing potential preference tags, use the real-time semantic analysis capabilities of the vertical large model, keenly perceive the subtle changes in the user's needs in a specific situation, and accurately extract the user's real-time preference tags at this moment, so that the recommendation system can adjust the recommendation direction in time and more accurately meet the user's immediate needs.

[0024] The vector library storage module has the function of efficiently storing and managing book feature vectors. It provides a structured and easy-to-retrieve storage space for the book tags and related information generated by the book feature extraction module, and builds a huge book feature vector library. This module supports fast vector retrieval and matching operations, can quickly respond to query requests from the tag matching module, and promptly return book-related information that matches the user's preferred tags, ensuring the efficient operation and fast response of the recommendation system.

[0025] The label matching module has the function of accurately matching user preferences with book features. It can use an efficient vector matching algorithm based on the user's real-time preference labels to quickly retrieve book labels with a high degree of matching in the vector library storage module, accurately recall books that meet the user's current needs, and extract detailed information of these books from the vector library, providing an accurate set of book candidates for generating personalized recommendation lists, ensuring the high relevance and accuracy of the recommendation results.

[0026] The book recommendation reading comprehension module has the function of generating personalized recommended book lists and recommendation words. It combines the user's real-time preference tags and the book-related information recalled by the tag matching module, and through the deep reading comprehension and text generation capabilities of the vertical large model, deeply analyzes the fit between user preferences and book content, explores the unique value and highlights of the book, and then generates a recommended book list and recommendation words that are both attractive and persuasive and meet the user's personalized needs, thereby improving the user's recognition and acceptance of the recommendation results and enhancing the recommendation effect.

[0027] The user follow-up module has the function of processing user follow-up questions and optimizing recommendation results. When the user has further requirements or is dissatisfied with the initially generated recommended book list, the module allows the user to express new requirements by asking follow-up questions. It can intelligently integrate the user's additional requirements with the original question, re-form an accurate description of the user's real question, and then reactivate the user's real-time preference extraction, label matching, and book recommendation reading comprehension modules, and regenerate personalized recommended book lists and recommendation words according to the optimized user needs. This module supports iterative questioning and recommendation until the user is completely satisfied with the recommendation results, which fully reflects the flexibility and user orientation of the recommendation system, and ensures that the recommendation service can continue to meet the user's ever-changing personalized needs.

[0028] Embodiment 2 of the present invention relates to a book recommendation method based on a vertical large model, which is implemented by the system of embodiment 1 and includes the following steps: Step S1: Based on the user's book list, user comments, and user search history information, a vertical large model is used to extract user features and generate user potential preference tags; specifically, the following steps are included: Step S11: Data collection and processing Collect user book lists: extract book titles, authors, and subject information from user book lists; Collect user reviews: Extract text content from user reviews of books; Collect user search history: record the keywords entered by the user in the search box and the search time; Text error correction: filter out typos, garbled characters, and irrelevant information; Step S12, using the vertical large model to extract user features, splicing the data collected and processed in step S11 into the vertical large model, and inputting the user feature extraction instruction "extracting user potential preferences according to user personalized information" into the vertical large model; Step S13: The self-developed vertical large model has the ability to extract user features, and will generate user potential preference labels, such as "historical novel lover" and "technology lover" according to the user feature extraction instruction in step S12; Step S2: Based on the book summary, book theme, book style, and book author information, the vertical large model is used to extract book features, generate book tags, and store relevant information in the vector library for subsequent book display; specifically, it includes: Step S21: Data collection Collect book summaries: Extract the summary part of the book.

[0029] Collect book topics: Extract category and topic tags of books.

[0030] Collect book styles: Extract the style description of the book, such as "humorous", "serious", etc.

[0031] Collect book author information: extract author's name, style, representative works and other information.

[0032] Step S22, using the vertical large model to extract book features, splicing the data collected in step S21 into the vertical large model, and inputting the book feature extraction instruction "extract book features according to book information" into the vertical large model; Step S23: The self-developed vertical large model has the ability to extract book features, and generates book labels such as "history" and "science fiction" according to the book feature extraction instructions in step S22; Step S24, vectorize the book tags and the book-related information in step S21 and store them in a vector library, which can be Elasticsearch or Faiss; the vectorization operation refers to converting text into mathematical vectors to facilitate subsequent similarity calculation; the vectorization method is to train a vectorization model based on own data and vectorize the book information through the vectorization model; Step S3: Combine the user's real questions and the user's potential preference tags, use the vertical large model to extract the user's real-time preference tags; Step S31, extract the user's real question and the user's potential preference label generated in step S1, splice the two and input them into the vertical large model, and at the same time input the user's real-time feature extraction instruction "extract the user's real thoughts based on the user's current question and the user's potential preference" into the vertical large model; Step S32: The self-developed vertical large model has the ability to extract user real-time preferences, and will generate user real-time preference tags according to the user real-time feature extraction instructions described in step S31, such as "currently interested in in-depth research on historical novels."

[0033] Step S4: Match the user's real-time preference tags with the book tags in the vector library, recall the books with a high matching degree, and return the relevant information of the books from the vector library; Step S41, vectorizing the user's real-time preference tags, the vectorization method is consistent with step S24; Step S42: Match the vectorized user real-time preference tags with the book tags in the vector library and use the cosine similarity between the two as the matching score.

[0034] Step S43: Recall the top 10 books with the highest matching degree according to the matching scores, and filter out the books that the user has read.

[0035] Step S44: extract detailed information of the books recalled in step S43 from the vector library, including the title, author, and abstract.

[0036] Step S5: Combine the user's real-time preference tags in step S3 and the book-related information recalled in step S4, use the vertical large model to perform reading comprehension, and generate a personalized recommended book list and personalized recommendation words; Step S51: Combine the user's real-time preference tag in step S3 and the book-related information recalled in step S4, splice the two and input them into the vertical large model, and at the same time input the personalized book list generation instruction "generate a personalized recommended book list based on the user's real-time preference and the most relevant book information" into the vertical large model; Step S52: The self-developed vertical large model has the ability to recommend books in a personalized manner, and generates a personalized book list according to the personalized book list generation instruction described in step S51; This step uses the deep reading comprehension capabilities of the vertical large model to explore the unique value and highlights of the book for users, generate attractive and persuasive recommendations, and enhance user recognition of the recommendation results.

[0037] Step S6: If the user has further requirements for the personalized recommended book list generated in step S5, the user can express the new requirements by asking questions, combining the user's additional requirements with the user's original questions, regenerating accurate real user questions, and re-going through the process of step S3-step S5 to regenerate the personalized recommended book list and personalized recommendation words based on the optimized user requirements. This step supports unlimited iterations of asking questions and recommendations. The system will continue to adjust the recommendation strategy based on user feedback until the user is completely satisfied with the recommendation results, which fully reflects the flexibility and user orientation of the recommendation system, ensuring that the recommendation service can continue to fit the user's changing personalized needs and provide users with the ultimate personalized book recommendation experience.

[0038] On the whole, the present invention is a self-developed vertical large model based on its own massive book data. It has deep reading comprehension capabilities and massive book knowledge. It can analyze the user's true intentions, generate personalized book lists, explore the unique value and highlights of books, and enhance user recognition of recommendation results.

[0039] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A book recommendation system based on a vertical large model, characterized in that: The system includes: a user feature extraction module, a book feature extraction module, a user real-time preference extraction module, a vector library storage module, a label matching module, a book recommendation reading comprehension module, and a user questioning module; The user feature extraction module accurately extracts key features from massive user behavior data, mines multi-dimensional information such as reading preferences, comment sentiment tendencies and focus points, and search history exploration intentions, and converts them into concise and accurate user potential preference labels; The book feature extraction module deeply analyzes and labels the multi-source information of the book, identifies the core content of the abstract, the subject knowledge field, the style, emotional color and artistic characteristics, and the key elements of the author's creative style, and converts them into clear and accurate book labels; The user real-time preference extraction module captures the user's current needs and preferences in real time, combines potential preference tags, keenly perceives changes in needs under specific circumstances, and accurately extracts real-time preference tags, so that the recommendation system can adjust the recommendation direction in a timely manner; The vector library storage module efficiently stores and manages book feature vectors, provides structured and easy-to-retrieve storage space for book tags and related information, builds a huge book feature vector library, and supports fast vector retrieval and matching operations; The tag matching module uses an efficient vector matching algorithm to quickly retrieve book tags with high matching degrees based on the user's real-time preference tags, recall books that meet the user's needs, extract detailed information, and provide an accurate book candidate set for generating a personalized recommended book list; The book recommendation reading comprehension module combines the user's real-time preference tags and recalled book information, analyzes the fit points through deep reading comprehension and text generation capabilities, explores the value highlights of the book, and generates attractive recommended book lists and recommendation words that meet personalized needs; The user inquiry module processes user inquiries and optimizes recommendation results, integrates additional requirements with original questions, reactivates related modules, regenerates recommended book lists and recommendation statements according to optimization requirements, and supports iterative inquiry and recommendation until the user is satisfied.

2. A book recommendation system based on a vertical large model according to claim 1, characterized in that: The vertical large model is trained based on massive data and can deeply understand the fuzzy needs of users and accurately locate their specific preferences; The vertical large model has deep language understanding capabilities and can support multiple tasks. It is used in the user feature extraction module, book feature extraction module, user real-time preference extraction module, book recommendation reading comprehension module, and user questioning module. According to the different inputs and instructions of each module, it generates results that comply with the instruction requirements, thereby serving multiple modules and providing strong technical support and core driving force for realizing personalized and precise book recommendations.

3. A book recommendation method based on a vertical large model, the method being applied to the book recommendation system based on a vertical large model as claimed in any one of claims 1 to 2, characterized in that: The method comprises the following steps: Step S1: Based on the user's book list, user comments, and user search history information, a vertical large model is used to extract user features and generate user potential preference tags; Step S2: Based on the book summary, book theme, book style, and book author information, the vertical large model is used to extract book features, generate book tags, and store relevant information in a vector library for subsequent book display; Step S3: Combine the user's real questions and the user's potential preference tags, use the vertical large model to extract the user's real-time preference tags; Step S4: Match the user's actual preference label with the book label in the vector library, recall the books with a high matching degree, and return the relevant information of the book from the vector library; Step S5: Combine the user's real-time preference tags in step S3 and the book-related information recalled in step S4, use the vertical large model to perform reading comprehension, and generate a personalized recommended book list and personalized recommendation words; Step S6: If the user has further requirements for the personalized recommended book list generated in step S5, the user can express the new requirements by asking follow-up questions, and the user's additional requirements are combined with the user's original questions to regenerate accurate real user questions. The process of steps S3 to S5 is repeated again to regenerate the personalized recommended book list and personalized recommendation words according to the optimized user requirements. This step supports unlimited iterations of questioning and recommendations. The system will continue to adjust the recommendation strategy based on user feedback until the user is completely satisfied with the recommendation result. This fully demonstrates the flexibility and user orientation of the recommendation system, ensuring that the recommendation service can continue to meet the user's ever-changing personalized needs and provide users with the ultimate personalized book recommendation experience.

4. The book recommendation method based on vertical large model according to claim 3 is characterized in that: The step S1 comprises: Step S11: Data collection and processing: Collect user book lists: extract book titles, authors, and subject information from user book lists; Collect user reviews: Extract text content from user reviews of books; Collect user search history: record the keywords entered by the user in the search box and the search time; Text error correction: filter out typos, garbled characters, and irrelevant information; Step S12, using the vertical large model to extract user features, splicing the data collected and processed in step S11 into the vertical large model, and at the same time inputting the user feature extraction instruction "extracting user potential preferences according to user personalized information" into the vertical large model; Step S13: The self-developed vertical large model has the ability to extract user features, and will generate user potential preference labels according to the user feature extraction instructions described in step S12.

5. The book recommendation method based on vertical large model according to claim 3 is characterized in that: The step S2 comprises: Step S21: Data collection: Collect book summaries: extract the summary part of the book; Collect book topics: extract book categories and topic tags; Collect book styles: extract the style description of books; Collect book author information: extract the author's name, style, and representative works; Step S22, using the vertical large model to extract book features, splicing the data collected in step S21 into the vertical large model, and inputting the book feature extraction instruction "extract book features according to book information" into the vertical large model; Step S23: The self-developed vertical large model has the ability to extract book features, and generates book tags according to the book feature extraction instructions in step S22; Step S24, vectorize the book tags and the book-related information in step S21 and store them in a vector library, which can be Elasticsearch or Faiss. The vectorization operation refers to converting text into mathematical vectors to facilitate subsequent similarity calculation. The vectorization method is to train a vectorization model based on own data and vectorize the book information through the vectorization model.

6. The book recommendation method based on vertical large model according to claim 3 is characterized in that: The step S3 comprises: Step S31, extract the user's real question and the user's potential preference label generated in step S1, splice the two and input them into the vertical large model, and at the same time input the user's real-time feature extraction instruction "extract the user's real thoughts based on the user's current question and the user's potential preference" into the vertical large model; Step S32: The self-developed vertical large model has the ability to extract user real-time preferences, and will generate user real-time preference tags according to the user real-time feature extraction instructions described in step S31.

7. The book recommendation method based on vertical large model according to claim 3 is characterized in that: The step S4 comprises: Step S41, vectorizing the user's real-time preference tags, the vectorization method is consistent with step S24; Step S42: Match the vectorized user real-time preference tags with the book tags in the vector library and use the cosine similarity between the two as the matching score; Step S43: recall the top 10 books with the highest matching degree according to the matching scores, wherein the books that the user has read are filtered out; Step S44: extract detailed information of the books recalled in step S43 from the vector library, including the title, author, and abstract.

8. The book recommendation method based on vertical large model according to claim 3 is characterized in that: The step S5 comprises: Step S51: Combine the user's real-time preference tag in step S3 and the book-related information recalled in step S4, splice the two and input them into the vertical large model, and at the same time input the personalized book list generation instruction "generate a personalized recommended book list based on the user's real-time preference and the most relevant book information" into the vertical large model; Step S52: The self-developed vertical large model has the ability to recommend books in a personalized manner, and will generate a personalized book list according to the personalized book list generation instruction described in step S51.

9. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is run, the device where the non-volatile storage medium is located is controlled to execute the book recommendation method based on the vertical large model as described in any one of claims 3-8.

10. An electronic device, characterized in that: It comprises a processor and a memory; the memory stores computer-readable instructions, and the processor is used to run the computer-readable instructions, wherein the computer-readable instructions, when running, execute the book recommendation method based on the vertical large model described in any one of claims 3-8.

Images