College library intelligent consultation method and system combining user portrait and sentiment analysis

Abstract

The invention discloses a university library intelligent consultation method and system combining a user portrait and sentiment analysis, and the method comprises the steps: collecting multi-source original data, and extracting a user structured attribute and an unstructured dialogue text; the user continuous dialogue sequence is coded, user behavior feature vectors are clustered to form a user behavior pattern library, and dynamic updating of a user portrait is achieved by calculating the similarity between a user new behavior and each clustering center; based on an adversarial network mechanism, taking a current dialogue text and a historical behavior context of a user as input, outputting sentiment polarity through a sentiment classifier, judging a user type through a user type discriminator, and optimizing a sentiment recognition process in combination with user type information; and receiving a user portrait, a real-time emotional state and a dialogue intention, carrying out key information screening on the historical dialogue of the user by adopting an attention mechanism, and generating a personalized reply according with the personality, the emotional state and the actual demand of the user by combining the current consultation context and utilizing a pre-trained language model.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and library information service technology, and in particular relates to an intelligent consultation method and system for university libraries that combines user profiling and sentiment analysis. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent consultation services have become an important part of university library services. Existing intelligent consultation systems are mostly based on predefined FAQ databases or retrieval-based dialogue models, lacking a deep understanding and response to users' personalized needs, particularly in emotion recognition, user intent understanding, and personalized response generation. Traditional methods fail to effectively utilize users' historical behavioral data, differences in emotional expression, and dynamic user profiles, resulting in rigid and impersonal consultation interactions that negatively impact user experience and service effectiveness.

[0003] Therefore, there is an urgent need to propose an intelligent consultation method and system for university libraries that combines user profiling and sentiment analysis. Summary of the Invention

[0004] To address the aforementioned technical issues, this invention proposes an intelligent consultation method and system for university libraries that combines user profiling and sentiment analysis, thereby enhancing the intelligence and personalization of library consultation services.

[0005] To achieve the above objectives, this invention provides an intelligent consultation method for university libraries that combines user profiling and sentiment analysis, comprising: Collect raw data from multiple sources and extract user structured attributes and unstructured dialogue text; The user's continuous dialogue sequence is encoded, the user behavior feature vector is clustered to form a user behavior pattern library, and the user profile is dynamically updated by calculating the similarity between the user's new behavior and each cluster center. Based on the adversarial network mechanism, the current dialogue text and historical behavior context of the user are used as input. The sentiment classifier outputs the sentiment polarity, the user type discriminator determines the user type, and the sentiment recognition process is optimized by combining the user type information. It receives user profiles, real-time emotional states, and dialogue intentions, uses an attention mechanism to filter key information from the user's historical dialogues, and combines this with the current consultation context to generate personalized responses that match the user's personality, emotional state, and actual needs using a pre-trained language model.

[0006] Optionally, extracting user structured attributes and unstructured dialogue text includes: Extract user identity type, college, grade, and major attributes using an information extraction script; A combination of rule-based and deep learning methods is used to annotate sentiment polarity and dialogue intent, and a multimodal user-dialogue sample set is constructed.

[0007] Optionally, dynamic updates to user profiles include: LSTM or Transformer models are used to model user consultation behavior and capture the dynamic changes in dialogue text; Utilize K-means or DBSCAN algorithms to build a user behavior pattern database and identify abnormal or significant changes in behavior.

[0008] Optionally, the user type discriminator may distinguish the following: Users are categorized as either professional users or general users. Professional users include teachers and graduate students, while general users include undergraduate students. The user type identification results are associated with the behavior types in the user profile.

[0009] Optionally, the training process for adversarial network mechanisms includes: Use a joint multi-task learning framework that includes a sentiment classifier and a user type discriminator; By using an adversarial training mechanism, the differences in emotional expression among different user groups can be captured, thereby improving the model's generalization ability in cross-group scenarios.

[0010] Optionally, the loss function of the joint multi-task learning framework can be designed as follows: ; in, and These represent the model's predictions for sentiment polarity and user type, respectively. For the total number of comments in the entire dataset, These represent the original sentiment and user type tags of the comment, respectively. It is the parameter set of the model. These are the parameters for L2 regularization.

[0011] Optionally, the generated response content includes: The generated responses can be adjusted in a controllable manner, regulating their formality, information density, and emotional tone; adapting to the preferences of different user groups.

[0012] On the other hand, to achieve the above objectives, the present invention also provides an intelligent consultation system for university libraries that combines user profiling and sentiment analysis, comprising: The data acquisition and preprocessing module is used to collect raw data from multiple sources and extract user structured attributes and unstructured dialogue text. The user profile building module is used to encode continuous dialogue sequences of users, cluster user behavior feature vectors to form a user behavior pattern library, and realize the dynamic updating of user profiles by calculating the similarity between new user behaviors and each cluster center. The emotion and intent recognition module is used to optimize the emotion recognition process based on the adversarial network mechanism, taking the user's current dialogue text and historical behavior context as input, outputting the emotion polarity through an emotion classifier, determining the user type through a user type discriminator, and combining user type information. The personalized response generation module receives user profiles, real-time emotional states, and dialogue intentions. It uses an attention mechanism to filter key information from the user's historical dialogues and combines this with the current consultation context to generate personalized responses that match the user's personality, emotional state, and actual needs using a pre-trained language model.

[0013] Technical effects of this invention: This invention discloses an intelligent consultation method and system for university libraries that combines user profiling and sentiment analysis. By constructing user profiles through dynamic dialogue data and combining them with adversarial neural networks, the accuracy of sentiment classification is significantly improved, demonstrating strong theoretical innovation and practical application value. The research method is systematic, the experimental design is rigorous, and the results provide technical support for subsequent personalized intelligent services. Attached Figure Description

[0014] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating the user profiling process according to an embodiment of the present invention. Figure 2 This is a flowchart illustrating an intelligent consultation method for university libraries that combines user profiling and sentiment analysis, according to an embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an intelligent consultation system for university libraries that combines user profiling and sentiment analysis, according to an embodiment of the present invention. Detailed Implementation

[0015] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0016] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0017] like Figure 2As shown, this embodiment provides an intelligent consultation method for university libraries that combines user profiling and sentiment analysis, including: Collect raw data from multiple sources and extract user structured attributes and unstructured dialogue text; The user's continuous dialogue sequence is encoded, the user behavior feature vector is clustered to form a user behavior pattern library, and the user profile is dynamically updated by calculating the similarity between the user's new behavior and each cluster center. Based on the adversarial network mechanism, the current dialogue text and historical behavior context of the user are used as input. The sentiment classifier outputs the sentiment polarity, the user type discriminator determines the user type, and the sentiment recognition process is optimized by combining the user type information. It receives user profiles, real-time emotional states, and dialogue intentions, uses an attention mechanism to filter key information from the user's historical dialogues, and combines this with the current consultation context to generate personalized responses that match the user's personality, emotional state, and actual needs using a pre-trained language model.

[0018] Furthermore, extracting user structured attributes and unstructured dialogue text includes: Extract user identity type, college, grade, and major attributes using an information extraction script; A combination of rule-based and deep learning methods is used to annotate sentiment polarity and dialogue intent, and a multimodal user-dialogue sample set is constructed.

[0019] Specifically, one implementation of this embodiment is as follows: Raw data was collected from multiple sources, including library consultation records, real-time dialogue logs, book borrowing history, seat reservation records, and electronic resource access behavior. Scripts were used to automatically extract structured user attributes (including identity type, college, grade, and major) and unstructured dialogue text. A combination of rule-based and deep learning methods was employed to label sentiment polarity and dialogue intent, constructing a high-quality, multimodal user-dialogue sample set to provide a data foundation for subsequent user profiling and sentiment analysis.

[0020] Furthermore, dynamic updates to user profiles include: LSTM or Transformer models are used to model user consultation behavior and capture the dynamic changes in dialogue text; Utilize K-means or DBSCAN algorithms to build a user behavior pattern database and identify abnormal or significant changes in behavior.

[0021] Specifically, such as Figure 1 One implementation of this embodiment is shown below: To achieve dynamic modeling of user consultation behavior, a time-series hierarchical model (such as LSTM or Transformer) is used to encode continuous user dialogue sequences, capturing their interest drift and behavioral evolution. Simultaneously, clustering algorithms (such as K-means or DBSCAN) are employed to cluster user behavior feature vectors, forming a library of typical user behavior patterns such as "research-oriented," "learning-oriented," and "consulting-oriented." By calculating the similarity between new user behaviors and the cluster centers in real time, abnormal or significant behavioral changes are identified, enabling dynamic updating and correction of user profiles.

[0022] Furthermore, the user type discriminator's discrimination includes: Users are categorized as either professional users or general users. Professional users include teachers and graduate students, while general users include undergraduate students. The user type identification results are associated with the behavior types in the user profile.

[0023] Furthermore, the training process for adversarial network mechanisms includes: Use a joint multi-task learning framework that includes a sentiment classifier and a user type discriminator; By using an adversarial training mechanism, the differences in emotional expression among different user groups can be captured, thereby improving the model's generalization ability in cross-group scenarios.

[0024] Specifically, one implementation of this embodiment is based on sentiment and user type recognition using adversarial networks. "User type" can be understood from two dimensions: identity type, such as teacher, graduate student, undergraduate student, etc., used to distinguish users' professional background and identity attributes; and behavior type, such as "research-oriented," "learning-oriented," "consulting-oriented," etc., used to describe users' typical behavioral patterns.

[0025] The "User Type Discriminator" in the Emotion and Intent Recognition module primarily identifies identity types, but its output can be correlated with behavior types in the user profile to further improve the accuracy of emotion recognition and response generation.

[0026] To improve the accuracy of sentiment and user type recognition, this module introduces a Generative Adversarial Network (GAN) mechanism on top of JOINT multi-task learning to further enhance generalization ability across user groups. Specifically, the system includes a sentiment classifier and a user type discriminator: Sentiment classifier: Takes the user's current dialogue text and historical behavior context as input, and outputs the sentiment polarity (positive / negative / neutral). User type discriminator: Determines whether a user is a professional user (such as a teacher or graduate student) or a general user (such as an undergraduate student), and optimizes the sentiment recognition process by combining user type information.

[0027] Furthermore, the loss function of the joint multi-task learning framework is designed as follows: ; in, and These represent the model's predictions for sentiment polarity and user type, respectively. For the total number of comments in the entire dataset, These represent the original sentiment and user type tags of the comment, respectively. It is the parameter set of the model. These are the parameters for L2 regularization. The AdaGrad gradient descent algorithm is used for optimization during model training, and the SkipGram algorithm is used in the experiments to generate word vectors.

[0028] Furthermore, the generated response content includes: The generated responses can be adjusted in a controllable manner, regulating their formality, information density, and emotional tone; adapting to the preferences of different user groups.

[0029] Specifically, one implementation of this embodiment involves receiving the user profile, real-time emotional state, and dialogue intent output by the preceding module. An attention mechanism is used to filter key information from the user's historical dialogues, combined with the current consultation context. A pre-trained language model (such as GPT, T5, or BERT) is then used to generate response content that matches the user's personality, emotional state, and actual needs. Furthermore, the system supports controllable adjustments to the generated responses, such as adjusting formality, information density, and emotional tone, to suit the preferences of different user groups.

[0030] Furthermore, the process of implementing the method in the intelligent consultation service platform includes: constructing a one-stop intelligent consultation service platform for university libraries, providing the following core functions: Intelligent Q&A and Service Guidance: Supports natural language questioning and multi-turn dialogue, providing consultation on library policies, service procedures, etc. Operational service implementation: Integrating functions such as document retrieval, borrowing inquiry, seat reservation, and event registration; Personalized resource recommendations: Proactively recommending books, academic resources, lecture information, etc., based on user profiles; Emotion-aware interaction: Identify the user's emotional state, adjust response strategies and tone, and enhance the interactive experience.

[0031] like Figure 3 As shown, this embodiment also provides an intelligent consultation system for university libraries that combines user profiling and sentiment analysis, including: The data acquisition and preprocessing module is used to collect raw data from multiple sources and extract user structured attributes and unstructured dialogue text. Furthermore, collect user consultation dialogue data and extract user attribute, sentiment, and intent information: User dialogue text and related behavioral data were collected from multiple sources, including the library's consultation system, chat logs, and borrowing records. User attributes (such as identity, college, and grade), sentiment polarity, and dialogue intent were extracted using information extraction scripts to construct a structured user-dialogue dataset.

[0032] The user profile building module is used to encode continuous dialogue sequences of users, cluster user behavior feature vectors to form a user behavior pattern library, and realize the dynamic updating of user profiles by calculating the similarity between new user behaviors and each cluster center. Furthermore, construct a dynamic user profile model: like Figure 1 As shown, a time-series hierarchical model is used to model user consultation behavior and capture the dynamic changes in dialogue text. Clustering algorithms (such as K-means) are used to cluster user behavior feature vectors to establish a user behavior pattern library. Abnormal or significant behaviors are identified by calculating the distance between new behaviors and cluster centers, and user profiles are dynamically updated.

[0033] The emotion and intent recognition module is used to optimize the emotion recognition process based on the adversarial network mechanism, taking the user's current dialogue text and historical behavior context as input, outputting the emotion polarity through an emotion classifier, determining the user type through a user type discriminator, and combining user type information. Furthermore, sentiment and user type recognition based on adversarial networks: A Generative Adversarial Network (GAN) structure is introduced to construct a sentiment classifier and a user type discriminator. The discriminator distinguishes whether a user is a professional user (such as a teacher or graduate student) or a general user (such as an undergraduate student), while the sentiment classifier incorporates user type information to enhance the accuracy of sentiment recognition. The loss function is designed as follows: ; in, and These represent the model's predictions for sentiment polarity and user type, respectively. For the total number of comments in the entire dataset, It refers to the original emotions and user type tags in the comments. It is the parameter set of the model. These are the parameters for L2 regularization to prevent overfitting; the AdaGrad gradient descent algorithm is used for optimization during model training, and the SkipGram algorithm is used in the experiment to generate word vectors.

[0034] By using adversarial training to capture the differences in emotional expression among different user groups, the model's generalization ability can be improved.

[0035] The personalized response generation module receives user profiles, real-time emotional states, and dialogue intentions. It uses an attention mechanism to filter key information from the user's historical dialogues and combines it with the current consultation context to generate personalized responses that match the user's personality, emotional state, and actual needs using a pre-trained language model. Furthermore, personalized responses can be generated by combining user profiles: By leveraging attention mechanisms to filter content relevant to the current inquiry from the user's historical dialogues, and combining user profiles, emotional states, and intent information, a pre-trained language model (such as GPT or BERT) is used to generate responses that match the user's personality and context.

[0036] Furthermore, the system also includes an intelligent consultation service platform module: This module integrates the above modules to develop an intelligent consultation service platform with the following functions: Personalized Q&A and professional consultation; Operational services such as document retrieval, borrowing inquiry, and seat reservation; Book and resource recommendations based on user profiles.

[0037] This invention discloses an intelligent consultation method and system for university libraries that combines user profiling and sentiment analysis. By constructing user profiles through dynamic dialogue data and combining them with adversarial neural networks, the accuracy of sentiment classification is significantly improved, demonstrating strong theoretical innovation and practical application value. The research method is systematic, the experimental design is rigorous, and the results provide technical support for subsequent personalized intelligent services.

[0038] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart consultation method for university libraries that combines user profiling and sentiment analysis, characterized in that: include: Collect raw data from multiple sources and extract user structured attributes and unstructured dialogue text; The user's continuous dialogue sequence is encoded, the user behavior feature vector is clustered to form a user behavior pattern library, and the user profile is dynamically updated by calculating the similarity between the user's new behavior and each cluster center. Based on the adversarial network mechanism, the current dialogue text and historical behavior context of the user are used as input. The sentiment classifier outputs the sentiment polarity, the user type discriminator determines the user type, and the sentiment recognition process is optimized by combining the user type information. It receives user profiles, real-time emotional states, and dialogue intentions, uses an attention mechanism to filter key information from the user's historical dialogues, and combines this with the current consultation context to generate personalized responses that match the user's personality, emotional state, and actual needs using a pre-trained language model.

2. The intelligent consultation method for university libraries combining user profiling and sentiment analysis as described in claim 1, characterized in that, Extracting user structured attributes and unstructured dialogue text includes: Extract user identity type, college, grade, and major attributes using an information extraction script; A combination of rule-based and deep learning methods is used to annotate sentiment polarity and dialogue intent, and a multimodal user-dialogue sample set is constructed.

3. The intelligent consultation method for university libraries combining user profiling and sentiment analysis as described in claim 1, characterized in that, Dynamic updates to user profiles include: LSTM or Transformer models are used to model user consultation behavior and capture the dynamic changes in dialogue text; Utilize K-means or DBSCAN algorithms to build a user behavior pattern database and identify abnormal or significant changes in behavior.

4. The intelligent consultation method for university libraries combining user profiling and sentiment analysis as described in claim 1, characterized in that, The user type discriminator's discrimination includes: Users are categorized as either professional users or general users. Professional users include teachers and graduate students, while general users include undergraduate students. The user type identification results are associated with the behavior types in the user profile.

5. The intelligent consultation method for university libraries combining user profiling and sentiment analysis as described in claim 1, characterized in that, The training process for adversarial network mechanisms includes: Use a joint multi-task learning framework that includes a sentiment classifier and a user type discriminator; By using an adversarial training mechanism, the differences in emotional expression among different user groups can be captured, thereby improving the model's generalization ability in cross-group scenarios.

6. The intelligent consultation method for university libraries combining user profiling and sentiment analysis as described in claim 5, characterized in that, The loss function of the joint multi-task learning framework is designed as follows: ； in, and These represent the model's predictions for sentiment polarity and user type, respectively. For the total number of comments in the entire dataset, These represent the original sentiment and user type tags of the comment, respectively. It is the parameter set of the model. These are the parameters for L2 regularization.

7. The intelligent consultation method for university libraries combining user profiling and sentiment analysis as described in claim 1, characterized in that, The generated response content includes: The generated responses can be adjusted in a controllable manner, regulating their formality, information density, and emotional tone; adapting to the preferences of different user groups.

8. A smart consultation system for university libraries that combines user profiling and sentiment analysis, characterized in that: For implementing the intelligent consultation method for university libraries that combines user profiling and sentiment analysis as described in any one of claims 1-7, the system comprises: The data acquisition and preprocessing module is used to collect raw data from multiple sources and extract user structured attributes and unstructured dialogue text. The user profile building module is used to encode continuous dialogue sequences of users, cluster user behavior feature vectors to form a user behavior pattern library, and realize the dynamic updating of user profiles by calculating the similarity between new user behaviors and each cluster center. The emotion and intent recognition module is used to optimize the emotion recognition process based on the adversarial network mechanism, taking the user's current dialogue text and historical behavior context as input, outputting the emotion polarity through an emotion classifier, determining the user type through a user type discriminator, and combining user type information. The personalized response generation module receives user profiles, real-time emotional states, and dialogue intentions. It uses an attention mechanism to filter key information from the user's historical dialogues and combines this with the current consultation context to generate personalized responses that match the user's personality, emotional state, and actual needs using a pre-trained language model.

Images