Method and device for carrying out personalized prediction on user based on CLLM model, and medium

Abstract

The invention relates to an artificial intelligence technology, and discloses a method for carrying out personalized prediction on a user based on a CLLM model, computer equipment and a storage medium, and the method comprises the steps: constructing a multi-dimensional data set according to collected user preference data and situation factor data; training a pre-constructed CLLM model based on the multi-dimensional data set, so that the CLLM model learns the relevance between the user preference data and the situation factor data, and has the capability of performing personalized output of a natural language based on the relevance; and utilizing the trained CLLM model to perform data analysis on the situation factor data related to the behavior of the target user, and predicting and generating a personalized result meeting the preference of the target user. The invention further discloses a control device. The invention aims to realize that the CLLM model accurately captures the user demand and improve the accuracy of personalized prediction of the user, so that the content output conforming to the user expectation is obtained through prediction.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, a control device, a computer device, and a computer-readable storage medium for performing personalized predictions on users based on a CLLM model. Background Art

[0002] With the rapid development of natural language processing technology, the CLLM (Contextualized Large Language Model) model has demonstrated excellent general processing capabilities in many fields. However, traditional model processing methods usually ignore the personalized needs and specific usage contexts of users (such as ignoring the differences between individual users and the changes in user behavior in different contexts), resulting in the content output by the model may deviate from the actual needs of users (for example, the model may not be able to respond to the user's current geographic location, time or environmental factors in real time, resulting in the lack of personalization and contextual relevance of the output content); in addition, existing models are also limited in integrating user preferences and contextual factors, usually focusing only on single-dimensional information analysis, without achieving deep integration of multi-dimensional data, and thus unable to fully capture the complexity of user behavior.

[0003] The above situations make it difficult for the CLLM model to accurately capture user needs, and it is also difficult to predict content output that meets user expectations, thereby reducing user satisfaction and trust in the model and affecting user experience of the model.

[0004] The above contents are only used to assist in understanding the technical solution of the present application and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of this application is to provide a method, control device, computer equipment and computer-readable storage medium for performing personalized predictions for users based on the CLLM model, aiming to enable the CLLM model to accurately capture user needs, improve the accuracy of personalized predictions for users, and thus predict content output that meets user expectations.

[0006] To achieve the above objectives, the present application provides a method for performing personalized prediction on users based on the CLLM model, comprising the following steps:

[0007] Constructing a multidimensional data set based on the collected user preference data and situational factor data; wherein the situational factor data is related to the user activity record, the user activity record is used to generate the user preference data, and the situational factor data includes at least one of the behavior time, the behavior location and the local weather;

[0008] The pre-built CLLM model is trained based on the multi-dimensional data set so that the CLLM model learns the correlation between user preference data and contextual factor data and has the ability to output personalized natural language based on the correlation;

[0009] Using the trained CLLM model, data analysis is performed on contextual factor data related to target user behavior, and personalized results that meet the preferences of target users are predicted and generated.

[0010] Optionally, before the step of constructing a multidimensional data set based on the collected user preference data and situational factor data, the step further includes:

[0011] Cleaning the user preference data and the situational factor data, removing duplicate data and erroneous data, and preprocessing the cleaned data;

[0012] The preprocessing method includes at least one of the following:

[0013] Unified data format;

[0014] Desensitize sensitive information;

[0015] Fill in missing values.

[0016] Optionally, the step of constructing a multidimensional data set based on the collected user preference data and situational factor data includes:

[0017] Creating a data mapping table based on predefined rules to associate user preference data with corresponding contextual factor data;

[0018] Select data storage technology and design data table structure, import user preference data and related situational factor data, and form a multidimensional data set.

[0019] Optionally, the method for performing personalized prediction for users based on the CLLM model further includes:

[0020] When training the CLLM model, the multidimensional dataset is divided into training set, validation set and test set;

[0021] When executing the main loop of model training, the training set is used for forward propagation and backward propagation, and the model parameters are updated;

[0022] After each training cycle, the validation set is used to evaluate the generalization ability of the model, monitor the training error and validation error, and generate cross-validation results;

[0023] Adjust the model's hyperparameters based on the cross-validation results;

[0024] After the model training is completed, the test set is used for test acceptance to evaluate the model training results.

[0025] Optionally, before the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the step further includes:

[0026] By using the data collection module, when the target user initiates a request, the contextual factor data related to the target user's behavior is automatically captured; and key features are extracted from the target user's historical behavior records to form historical data;

[0027] Preprocess the situational factor data and historical data corresponding to the target users so that the processed data meets the input requirements of the CLLM model;

[0028] The situational factor data and historical data corresponding to the target users are input into the trained CLLM model.

[0029] Optionally, after the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the method further includes:

[0030] Obtain feedback from target users regarding personalized results;

[0031] Conduct semantic analysis on feedback information, mark feedback types and relevance, and generate feedback data;

[0032] The feedback data is associated with the data related to the personalized result, and the CLLM model is updated based on the associated result.

[0033] Optionally, after the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the method further includes:

[0034] Establish a model update cycle to retrain the model regularly; and,

[0035] An online learning mechanism is introduced to make instant adjustments to the model using real-time data streams.

[0036] To achieve the above object, the present application also provides a control device, comprising:

[0037] A construction module, configured to construct a multidimensional data set based on the collected user preference data and situational factor data; wherein the situational factor data is related to the user activity record, the user activity record is used to generate the user preference data, and the situational factor data includes at least one of the behavior time, the behavior location and the local weather;

[0038] A training module, used to train the pre-built CLLM model based on the multi-dimensional data set, so that the CLLM model learns the correlation between the user preference data and the situational factor data, and has the ability to output personalized natural language based on the correlation;

[0039] The prediction module is used to use the trained CLLM model to perform data analysis on contextual factor data related to the target user's behavior, and predict and generate personalized results that meet the target user's preferences.

[0040] To achieve the above objectives, the present application also provides a computer device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, the steps of the method for performing personalized predictions for users based on the CLLM model as described above are implemented.

[0041] To achieve the above objectives, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method for performing personalized prediction for users based on the CLLM model as described above are implemented.

[0042] The method, control device, computer equipment and computer-readable storage medium for personalized prediction of users based on the CLLM model provided in this application enable the CLLM model to accurately capture user needs, provide highly personalized prediction results, and generate natural and fluent language output through a deep understanding of user preferences and situational factors, so as to improve the accuracy of personalized predictions for users, thereby predicting content output that meets user expectations and improving user satisfaction, trust and usage experience with the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a schematic diagram of the steps of a method for performing personalized prediction on a user based on the CLLM model in one embodiment of the present application;

[0044] Figure 2 This is a schematic diagram of a multidimensional data set construction process in an embodiment of the present application;

[0045] Figure 3 This is a schematic diagram of a model generation process in an embodiment of the present application;

[0046] Figure 4 This is a schematic diagram of the model application process in one embodiment of the present application;

[0047] Figure 5 This is a schematic diagram of a control device in an embodiment of the present application;

[0048] Figure 6 A schematic diagram of the internal structure of a computer device according to an embodiment of the present application.

[0049] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0050] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0051] In addition, if the description of "first", "second", etc. is involved in this application, it is only used for descriptive purposes (such as for distinguishing the same or similar features), and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in this field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such combination of technical solutions does not exist and is not within the scope of protection required by this application.

[0052] Reference Figure 1 In one embodiment, a method for performing personalized prediction on a user based on a CLLM model includes:

[0053] Step S10: construct a multidimensional data set based on the collected user preference data and situational factor data; wherein the situational factor data is related to the user activity record, the user activity record is used to generate the user preference data, and the situational factor data includes at least one of the behavior time, the behavior location and the local weather;

[0054] Step S20: training the pre-built CLLM model based on the multidimensional data set, so that the CLLM model learns the correlation between the user preference data and the situational factor data, and has the ability to perform personalized output in natural language based on the correlation;

[0055] Step S30: Utilize the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior, and predict and generate personalized results that meet the target user's preferences.

[0056] In this embodiment, the execution terminal of the embodiment may be a computer device, or may be other equipment or devices (such as a control device) that controls the computer device.

[0057] As described in step S10, user preference data may include information such as user interests, historical behaviors, interaction records, etc. These data help build user portraits and understand users' long-term preferences.

[0058] Situational factor data can be the time of behavior, location of behavior, and local weather, etc. These data provide the user's situational context at a specific moment and help understand the user's needs in different environments.

[0059] Integrating user preference data and situational factor data into a multidimensional data set can fully capture the complexity of user behavior and the impact of the environment.

[0060] Optionally, by comprehensively collecting user preference data (such as historical behavior records, social media activities, questionnaire survey results, etc.) and situational factor data (such as time, location, weather conditions, etc.), a multidimensional data set of user behavior and situational information is constructed to lay the foundation for subsequent personalized analysis.

[0061] In the process of data preparation and multi-source data fusion, ensure that the collected user preference data and situational factor data are detailed, accurate and timely, providing a solid foundation for subsequent analysis and model training. (The specific collection method and dimension are optional according to actual conditions).

[0062] Optionally, user preference data can be generated based on user activity records:

[0063] Search engine log analysis: Automatically capture the user's query keywords in the search engine, analyze the search frequency, trends and topic distribution to identify the user's areas of interest;

[0064] Purchase history analysis: Collect users’ purchase records from e-commerce platforms, including product categories, price ranges, and purchase frequencies, to gain insights into users’ consumption preferences;

[0065] Browsing behavior tracking: monitor users’ browsing trajectory on the website, including dwell time, click-through rate and page jump pattern, to analyze users’ reading preferences and attention focus;

[0066] Social media activity aggregation: Capture users’ posts, comments, and sharing records on social media platforms, and analyze their sentiment tendencies and engagement with hot topics through natural language processing technology;

[0067] Questionnaires and feedback: Design and distribute questionnaires covering interests, hobbies, lifestyle habits, product preferences, etc., and collect direct feedback from users to obtain more in-depth personalized information.

[0068] Optionally, contextual factors data can be collected based on relevant data recorded from user activity:

[0069] Timestamp recording: Record precise timestamps for each user action, including date, hour, minute, and even second, to analyze the impact of time on user behavior;

[0070] Geolocation tracking: Through user-authorized GPS data, the user's specific location information is recorded, including latitude and longitude coordinates, and possible location types (such as home, office, shopping mall, etc.);

[0071] Weather information capture: Integrate third-party meteorological APIs to obtain real-time weather conditions in the user's location, including temperature, humidity, wind level, and weather type, to assess the potential impact of weather on user activities.

[0072] Optional, see Figure 2 ,After collecting user preference data and situational factor data, data can be cleaned and fused to build a unified multidimensional data set to ensure the relevance, integrity and consistency of the data, and provide an optimized data structure for the subsequent deep learning model training.

[0073] As described in step S20, based on the collected multi-source data, deep learning techniques (such as convolutional neural networks, LSTM (Long Short-Term Memory) networks, self-attention mechanisms, etc.) are used to build a CLLM model that can capture the complex relationship between user preferences and situational factors. This model can not only learn the user's historical behavior patterns, but also perceive situational changes in real time, and achieve a deep integration of personalized needs and dynamic situations.

[0074] Optionally, a CLLM model is pre-built based on deep learning technology, so that it can effectively capture the time series characteristics of user behavior, while integrating user preferences and situational factors to generate highly personalized outputs.

[0075] Optional, model architecture selection: Plan the structure of the LSTM layer in detail, including the number of layers, number of nodes, and activation function, to ensure that the model can effectively process time series data.

[0076] Optional, self-attention mechanism integration: Add a self-attention module to the LSTM model and design an attention weight calculation formula to enhance the model's ability to capture key information, especially when processing long sequence data.

[0077] Optional, multi-input fusion design: Build a multi-input interface that allows the model to receive both user preference data and contextual factor data, and use appropriate feature fusion strategies such as concatenation, weighted sum, or gating mechanism to ensure effective integration of the two types of data.

[0078] Optional, output layer design: Define the output layer structure of the model, including output dimension and activation function, to ensure that the model can generate personalized output that meets the requirements.

[0079] The CLLM model can effectively learn the complex correlation between user preference data and contextual factor data by combining LSTM networks, self-attention mechanisms, and multi-input fusion strategies. Through training on multidimensional data sets, the model can generate highly personalized natural language output to meet the specific needs and context requirements of users.

[0080] As described in step S30, after the model training is completed, the trained CLLM model is used to perform deep preprocessing on the input data of the target user, and data analysis is performed on the contextual factor data related to the target user's behavior to generate highly personalized language output that closely fits the current context, that is, to predict and generate personalized results that meet the preferences of the target user.

[0081] This process ensures that the output content of the CLLM model not only reflects the personalized preferences of the target users, but also fully considers the impact of the real-time context, thereby greatly improving the relevance and attractiveness of the output content.

[0082] For example, suppose you have trained a CLLM model, such as a model that uses a combination of a convolutional neural network and an LSTM network, to predict a user's interest in movies. The output of the model is a predicted score for each movie, ranging from 1 to 5, where 5 indicates extremely high interest and 1 indicates almost no interest. To turn these predicted scores into a list of recommendations, you can use the following algorithm:

[0083] (1) Sorting algorithm: First, all movies are sorted according to the predicted ratings, from high to low.

[0084] (2) Filtering algorithm: Remove movies that the user has already watched to ensure that the recommendations are content that the user has not yet experienced.

[0085] (3) Diversity algorithm: In order to provide diverse recommendations, a diversity algorithm can be used to ensure that the recommendation list contains different types of movies and avoid recommending overly homogeneous content.

[0086] Based on the above algorithm, generate a recommendation list of 10 movies, each with a predicted rating and a short description. For each movie, provide a short personalized recommendation reason, such as "Based on your historical preferences, we think you will like the plot and actors of this movie." Create a card for each movie, showing the movie poster, title, predicted rating, and recommendation reason. Allow users to provide feedback on the recommendation, such as "I like it" and "Don't like it" buttons, to further personalize the recommendation.

[0087] Optionally, based on innovative multi-source data fusion strategies, users' historical behavior data, social media activities, questionnaire results, and contextual factors such as time, location, and weather are integrated. The construction of the deep learning model uses advanced CNN, LSTM, and self-attention mechanisms to capture the complex relationship between user preferences and contextual factors. Personalized output generation uses the trained model for deep preprocessing to generate language output that fits the current context. The continuous learning and optimization mechanism collects user feedback and the latest behavior data regularly, continuously adjusts the model parameters, and realizes the self-evolution of the model.

[0088] In one embodiment, by deeply understanding user preferences and contextual factors, the CLLM model accurately captures user needs, provides highly personalized prediction results, and generates natural and fluent language output to improve the accuracy of personalized predictions for users, thereby predicting content output that meets user expectations and improving user satisfaction, trust, and usage experience with the model. These advantages give the CLLM model significant advantages in improving user experience and meeting personalized needs, and it demonstrates excellent adaptability and robustness in multiple application scenarios.

[0089] The CLLM model trained in this way can more accurately understand and respond to users' personalized needs and situational changes, significantly improving the model's output quality and user experience, and providing more accurate technical support for intelligent customer service, personalized recommendations and other fields. Through the deep integration of user preferences and situational factors, the model output is highly personalized and situationally adaptable, which has important theoretical value and broad application prospects.

[0090] In one embodiment, based on the above embodiment, before the step of constructing a multidimensional data set based on the collected user preference data and context factor data, the step further includes:

[0091] Cleaning the user preference data and the situational factor data, removing duplicate data and erroneous data, and preprocessing the cleaned data;

[0092] The preprocessing method includes at least one of the following:

[0093] Unified data format;

[0094] Desensitize sensitive information;

[0095] Fill in missing values.

[0096] In this embodiment, in order to construct a multidimensional data set based on the collected user preference data and situational factor data, data cleaning and preprocessing are required after data collection.

[0097] Optionally, use a data cleaning algorithm to remove duplicate records and obviously erroneous data points to ensure data purity.

[0098] Optionally, unify data formats, such as converting date and time into ISO standard formats and encoding text data to ensure that all data can be correctly parsed and processed under a unified framework.

[0099] Optional, sensitive information desensitization - use hash encryption, fuzzy matching or other security measures to anonymize sensitive information such as user name, address, phone number, etc. to protect user privacy.

[0100] Optional, missing value filling: For a small amount of missing data, statistical filling (such as mean, mode), predictive filling, or pre- and post-value interpolation can be used to make reasonable estimates.

[0101] Data cleaning and preprocessing are the basis for building high-quality multidimensional data sets. By removing duplicate and erroneous data, unifying data formats, desensitizing data, and filling in missing values, we can ensure the accuracy, consistency, and security of the data, laying a good foundation for subsequent model training and analysis.

[0102] In one embodiment, based on the above embodiment, the step of constructing a multidimensional data set according to the collected user preference data and context factor data includes:

[0103] Creating a data mapping table based on predefined rules to associate user preference data with corresponding contextual factor data;

[0104] Select data storage technology and design data table structure, import user preference data and related situational factor data, and form a multidimensional data set.

[0105] In this embodiment, mapping rules for predefined data (i.e., predefined rules) are formulated to clarify the correspondence between user preference data and context factor data, such as how a user ID should be associated with a timestamp, geographic location information, etc.

[0106] In order to ensure the effective association between user preference data and context factor data and to achieve accurate provision of personalized services, the following specific and detailed predefined rules may be formulated in this embodiment (these rules are intended to guide the data collection, processing and integration process and ensure the consistency, accuracy and completeness of the data):

[0107] (1) Mapping between user ID and basic attributes: Each user ID will be associated with a series of basic attributes, including but not limited to age, gender, occupation, hobbies, etc. These attributes should be collected when the user registers and updated when the user completes the personal information. Use a globally unique identifier to ensure that each user has a unique identification code. Use structured database storage, and establish a foreign key association between each attribute field and the user ID field to ensure data consistency and reference integrity.

[0108] (2) Association between timestamps and user activities: All user activity records should contain timestamps to track the temporal sequence of user behavior and facilitate analysis of user behavior patterns in different time periods. The ISO 8601 standard (such as 2023-04-05T14:30:00Z) can be used uniformly to automatically record the operation type, operation time, user ID, and related context information for each user operation.

[0109] (3) Linking of geolocation information: User geolocation information should be linked to user activity records in order to analyze user behavior preferences in specific locations. The geolocation accuracy should be at least accurate to the city level, and if the device allows, it can be more accurate to the street or building level. Ensure that the collection of geolocation information complies with local laws and regulations, obtain explicit consent from users, and provide privacy setting options.

[0110] (4) Comprehensive consideration of contextual factors: Combine user preferences with contextual factors (such as weather, holidays, social events, etc.) to build a multi-dimensional user profile to support context-aware service recommendations. Among them, weather information obtains real-time weather conditions in the user's area through API calls, including temperature, humidity, wind speed, etc., maintains a dynamic calendar system, tracks important global holidays and local social events, and associates them with user locations and timestamps.

[0111] (5) Data integration and standardization: Ensure that all collected data is preprocessed and standardized to eliminate data inconsistencies and improve the efficiency and accuracy of data analysis. Data cleaning removes duplicates, outliers, and missing values ​​to ensure data quality; format standardization unifies the naming, units, and formats of data fields to facilitate cross-system data integration and query.

[0112] (6) Security and Compliance: During the entire data processing process, we comply with data protection regulations, take encryption, desensitization and other measures to protect user privacy, and ensure the secure storage and transmission of data. Among them, data encryption uses industry-standard encryption algorithms (such as AES) to encrypt and store sensitive data. Implement strict permission management, and only authorized personnel can access sensitive data.

[0113] Through the formulation and implementation of the above predefined rules, user preference data can be effectively associated with situational factor data to provide users with a more personalized and intelligent service experience. At the same time, the standardization, security and legality of the data processing process can be ensured to safeguard user rights and interests and the compliance of business operations.

[0114] Optionally, based on predefined rules, a data mapping table can be created using database technology or data processing tools to ensure that each piece of user preference data can be accurately linked to the corresponding contextual factor data. After the mapping table is created, a data association test can be performed to check whether all associated data points can be correctly matched to avoid data quality problems caused by association errors.

[0115] Optionally, the cube building process is as follows:

[0116] (1) Select data storage technology: Choose a suitable data warehouse or database management system, such as Hadoop HDFS, Amazon Redshift, Google BigQuery, etc., based on the data volume, access frequency, and security requirements;

[0117] (2) Designing the data table structure: Designing the table structure of the multidimensional dataset to ensure that it can accommodate all necessary fields of user preference data and contextual factor data, such as user ID, preference type, preference strength, timestamp, geographic location coordinates, etc.

[0118] (3) Data import and verification: Import the pre-processed user preference data and situational factor data into the designed data table, use scripts or tools to import the data, and perform preliminary data verification to ensure the correctness of the data import;

[0119] (4) Build data index: To speed up data query and analysis, build indexes for key fields such as user ID and timestamp to improve data retrieval efficiency.

[0120] Optionally, after the multidimensional dataset is built, further data quality checks can be performed, such as:

[0121] Integrity check: Check for null values ​​or missing data through SQL queries or data processing scripts to ensure that all fields are fully populated.

[0122] Consistency check: Compare the data mapping table with the multidimensional data set to confirm the consistency of the data and avoid inconsistent data records.

[0123] Correlation verification: Verify the correlation between user preference data and situational factor data again to ensure that the logical relationship between the data is correct.

[0124] Data verification and correction: Based on the above inspection results, correct the data quality problems found, such as supplementing missing values, correcting erroneous values, or deleting outliers, until the data set meets the predetermined quality standards.

[0125] Through the above solution, we can ensure the deep integration of user preference data and situational factor data to form a high-quality multidimensional data set, providing a solid data foundation for the subsequent construction of deep learning models, and then supporting the accurate recommendation and optimization of personalized services.

[0126] In one embodiment, based on the above embodiment, the method for performing personalized prediction on users based on the CLLM model further includes:

[0127] When training the CLLM model, the multidimensional dataset is divided into training set, validation set and test set;

[0128] When executing the main loop of model training, the training set is used for forward propagation and backward propagation, and the model parameters are updated;

[0129] After each training cycle, the validation set is used to evaluate the generalization ability of the model, monitor the training error and validation error, and generate cross-validation results;

[0130] Adjust the model's hyperparameters based on the cross-validation results;

[0131] After the model training is completed, the test set is used for test acceptance to evaluate the model training results.

[0132] In this embodiment, refer to Figure 3 When building the model, based on the LSTM architecture, self-attention mechanism, multi-input fusion and output layer design, after completing the model construction, we begin to prepare for model training.

[0133] Dataset division: The fused multidimensional dataset is divided into training set, validation set and test set, with the optional ratios of 70%, 15%, and 15% to ensure the randomness and representativeness of the data.

[0134] Parameter initialization: Set the initial learning rate, loss function, and optimizer to assign reasonable initial values ​​to the model parameters.

[0135] Training loop: The main loop that executes model training, uses the training set data for forward propagation and backpropagation, and updates the model parameters until a predetermined training round (epoch) is reached or the stopping condition is met.

[0136] Cross-validation: After each training cycle, use the validation set to evaluate the generalization ability of the model and monitor the training error and validation error to prevent overfitting.

[0137] Hyperparameter adjustment: According to the results of cross-validation, adjust the model's hyperparameters, such as learning rate, batch size, regularization coefficient, etc., to optimize model performance.

[0138] Model saving: During the training process, the model state, including weights and biases, is saved periodically so that training can be resumed or subsequent model evaluation can be performed when needed.

[0139] After the model training is completed, the final model is evaluated using the test set. The performance indicators of the model on the test set, such as accuracy, recall, F1-score, and AUC, are calculated to determine the performance of the model on unseen data. The test results are analyzed to evaluate whether the model has achieved the expected results and verify the practical application capabilities of the model.

[0140] In one embodiment, by dividing the multidimensional data set into a training set, a validation set, and a test set, and performing model training, evaluation, and adjustment through the above process, the accuracy and generalization ability of the CLLM model in personalized prediction tasks can be effectively improved. This ensures the robustness of the model and provides reliable support for practical applications.

[0141] In one embodiment, based on the above embodiment, before the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the step further includes:

[0142] By using the data collection module, when the target user initiates a request, the contextual factor data related to the target user's behavior is automatically captured; and key features are extracted from the target user's historical behavior records to form historical data;

[0143] Preprocess the situational factor data and historical data corresponding to the target users so that the processed data meets the input requirements of the CLLM model;

[0144] The situational factor data and historical data corresponding to the target users are input into the trained CLLM model.

[0145] In this embodiment, refer to Figure 4,When the model is applied, real-time situational information can be collected for the target user to extract the ,corresponding situational factor data; and key features can be extracted from the ,historical behavior records of the target user to form historical data.

[0146] Among them, real-time situational information collection: develop a data collection module to automatically capture the current situational information of the target user, such as geographic location, weather conditions, timestamp, etc. when the target user initiates a request.

[0147] Historical data integration: Establish a data integration system to extract key features from the historical behavior records of target users, including but not limited to past interaction patterns, preference settings, commonly used functions, etc.

[0148] Data preprocessing: Perform data preprocessing on the contextual factor data and historical data corresponding to the target users, implement data cleaning and preprocessing processes to ensure that the quality and format of all input data are consistent and suitable for model input requirements.

[0149] The processed situational factor data and historical data are used as input, and the trained CLLM model is called to perform prediction tasks and generate personalized outputs.

[0150] Optionally, design an output parsing algorithm to convert the raw output generated by the model into a user-understandable form, such as natural language text, recommendation lists, etc., and display it to the user through the user interface.

[0151] In the output parsing and presentation phase, the goal is to convert the raw output of the CLLM model into a format that users can intuitively understand and use. This process includes but is not limited to interpreting model predictions, generating understandable feedback, and designing a user-friendly interface to display this information (i.e., user interface presentation).

[0152] In some optional embodiments, after the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the method further includes:

[0153] Obtain feedback from target users regarding personalized results;

[0154] Conduct semantic analysis on feedback information, mark feedback types and relevance, and generate feedback data;

[0155] The feedback data is associated with the data related to the personalized result, and the CLLM model is updated based on the associated result.

[0156] Among them, user feedback: after outputting personalized results to target users, user feedback can also be monitored to obtain feedback information from target users on personalized results. For example, a user behavior tracking system can be integrated to monitor user feedback on generated content in real time, including but not limited to indicators such as click-through rate, dwell time, and conversion rate; user feedback channels can also be set up to collect users' direct evaluation of personalized outputs, distinguish between positive and negative feedback, and store them in categories.

[0157] Feedback evaluation: Implement the feedback analysis process, perform semantic analysis on the collected user feedback, annotate the feedback type (such as satisfaction, improvement suggestions) and relevance (feedback information can be weighted or scored based on the correlation between user feedback and predicted results, so that highly relevant information can be given priority in subsequent processing), generate feedback data, and associate the feedback data with the generated personalized results (such as recommended items, personalized content).

[0158] For example, the correspondence between feedback information and personalized results is updated in the database, such as creating a table to record each user's feedback and corresponding recommendation information for statistical analysis and model updating.

[0159] Optionally, the CLLM model can be retrained or fine-tuned using new information obtained from user feedback to better adapt to the user's real needs and changes in preferences. In model training, a reinforcement learning strategy can be used to train the model to improve the accuracy of personalized recommendations by taking user feedback as a reward signal. For example, if the user is satisfied with the recommendation result, the weight of the corresponding feature can be increased; otherwise, it can be reduced.

[0160] Model optimization: Regularly integrate new feedback data and user behavior information into the training data set to keep the data set timely and comprehensive. Execute the model update process, retrain the model using the latest data set, and adjust the network parameters and structure to adapt to changes in user needs.

[0161] After retraining the model, fine-tune the parameters, optimize the hyperparameter configuration, and continuously monitor the model performance indicators to ensure the steady improvement of output quality. In addition, A / B testing can be implemented to compare the output effects of the new and old models, and the actual benefits of the model update can be verified through experiments to ensure the effectiveness of the improvement measures.

[0162] By continuously collecting and analyzing user feedback and optimizing the model, a virtuous feedback loop is formed, which will enhance the adaptability of the model, improve user satisfaction, and promote the continuous evolution of personalized recommendations.

[0163] In one embodiment, by obtaining user feedback, performing semantic analysis, annotating feedback types, generating structured feedback data, and associating the feedback data with personalized results, the CLLM model can be effectively updated and optimized to more accurately meet the preferences of target users. This feedback mechanism not only improves the personalized service capabilities of the model, but also enhances the user experience.

[0164] In one embodiment, based on the above embodiment, after the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the method further includes:

[0165] Establish a model update cycle to retrain the model regularly; and,

[0166] An online learning mechanism is introduced to make instant adjustments to the model using real-time data streams.

[0167] In this embodiment, a model update cycle is established to ensure that the model is retrained regularly to reflect new changes in user preferences and situations. An online learning mechanism is introduced to enable the model to be adjusted instantly according to real-time data streams, improving response speed and accuracy.

[0168] Optionally, according to business needs and user feedback frequency, set the time interval for model updates, such as weekly, monthly or quarterly updates. This time interval should match the frequency of data changes to ensure the model's sensitivity to new data. You can also set certain conditions to trigger model updates, such as when user feedback data reaches a certain amount, user preferences change significantly, or when the performance of the validation set decreases.

[0169] Optionally, after each model update, record the model's training and validation performance to ensure that the new model performs better than the old model in actual applications. In addition, version control is performed on each update to facilitate rollback to the previous version when problems occur, ensuring the stability of the system.

[0170] Optionally, an online learning mechanism is introduced, and an algorithm suitable for online learning is selected, such as SGD (stochastic gradient descent) based on small batch data updates or other algorithms suitable for dynamic updates. These algorithms can adjust model parameters in real time after receiving new data without completely retraining the model.

[0171] Build a real-time data stream access mechanism to collect real-time user feedback and behavior data through APIs or message queues (such as Kafka, RabbitMQ, etc.) to ensure the timeliness and integrity of the data stream. Regularly update the model in small batches based on the data received in real time. For example, after receiving a certain amount of new user feedback, use this data to fine-tune the model to ensure that the model can quickly adapt to new user needs and changes.

[0172] Optionally, establish a monitoring system to monitor the performance of the model and user feedback in real time. Evaluate the model's prediction accuracy by comparing the model's prediction results with the actual performance. When the model performance degrades or becomes abnormal, an alarm is automatically triggered for manual inspection and intervention.

[0173] As user preferences and behavior patterns change, regularly evaluate the relevance and importance of existing features, add new features or remove invalid features as appropriate, and improve the performance of the model.

[0174] Optionally, during the online learning process, user feedback can be integrated in real time to evaluate the effectiveness of predictions and user satisfaction, providing a basis for model updates. Through continuous model updates and online learning mechanisms, a closed loop of data collection-feedback-model adjustment is formed to continuously improve the accuracy of personalized services.

[0175] In one embodiment, by establishing a periodic model update mechanism and introducing an online learning mechanism, it is possible to effectively respond to changes in user preferences and the dynamics of the data environment, ensuring that the CLLM model always performs well. This not only enhances the flexibility and adaptability of the model, but also improves the real-time nature of personalized recommendations and user satisfaction, providing strong support for continuous optimization.

[0176] In addition, refer to Figure 5 In an embodiment of the present application, a control device Z10 is further provided, comprising:

[0177] A construction module Z11 is used to construct a multidimensional data set based on the collected user preference data and situational factor data; wherein the situational factor data is related to the user activity record, the user activity record is used to generate the user preference data, and the situational factor data includes at least one of the behavior time, the behavior location and the local weather;

[0178] A training module Z12 is used to train a pre-built CLLM model based on a multi-dimensional data set, so that the CLLM model learns the correlation between user preference data and situational factor data, and has the ability to perform personalized output in natural language based on the correlation;

[0179] The prediction module Z13 is used to use the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior, and predict and generate personalized results that meet the preferences of the target user.

[0180] Optionally, the control device Z10 may be a virtual control device (such as a virtual machine) or a physical device (such as a physical device other than a computer device that can execute the corresponding method).

[0181] In addition, a computer device is also provided in an embodiment of the present application. The internal architecture of the computer device can be as follows: Figure 6 As shown, it includes a processor, a memory, a communication interface and an input interface connected by a system bus. The processor is used to provide computing and control capabilities. The memory 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 is used to store data called by the computer program. The communication interface is used to communicate data with an external terminal. The input interface is used to receive a signal input by an external device. When the computer program is executed by the processor, a method for personalized prediction of users based on the CLLM model as described in the above embodiment is implemented.

[0182] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the present application scheme, and does not constitute a limitation on the computer device to which the present application scheme is applied. For example, in some optional embodiments, the computer device may also include an output interface (not shown in the figure), and the output interface is also connected to the system bus and is used to output corresponding signals to the peripheral device.

[0183] In addition, the present application also proposes a computer-readable storage medium, the computer-readable storage medium includes a computer program, and when the computer program is executed by a processor, the steps of the method for performing personalized prediction on users based on the CLLM model as described in the above embodiment are implemented. It can be understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0184] In summary, the method, control device, computer device and computer-readable storage medium for personalized prediction of users based on the CLLM model provided in the embodiments of the present application enable the CLLM model to accurately capture user needs, provide highly personalized prediction results, and generate natural and fluent language output by deeply understanding user preferences and situational factors, so as to improve the accuracy of personalized prediction of users, thereby predicting content output that meets user expectations and improving user satisfaction, trust and usage experience of the model. These advantages make the CLLM model have significant advantages in improving user experience and meeting personalized needs, and show excellent adaptability and robustness in multiple application scenarios.

[0185] The CLLM model trained in this way can more accurately understand and respond to users' personalized needs and situational changes, significantly improving the model's output quality and user experience, and providing more accurate technical support for intelligent customer service, personalized recommendations and other fields. Through the deep integration of user preferences and situational factors, the model output is highly personalized and situationally adaptable, which has important theoretical value and broad application prospects.

[0186] Those of ordinary skill 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 memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0187] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0188] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for personalized prediction of users based on the CLLM model, characterized in that: include: Constructing a multidimensional data set based on the collected user preference data and situational factor data; wherein the situational factor data is related to the user activity record, the user activity record is used to generate the user preference data, and the situational factor data includes at least one of the behavior time, the behavior location and the local weather; The pre-built CLLM model is trained based on the multi-dimensional data set so that the CLLM model learns the correlation between user preference data and contextual factor data and has the ability to output personalized natural language based on the correlation; Using the trained CLLM model, data analysis is performed on contextual factor data related to target user behavior, and personalized results that meet the target user's preferences are predicted and generated.

2. The method for performing personalized prediction on users based on the CLLM model according to claim 1, characterized in that: Before the step of constructing a multidimensional data set based on the collected user preference data and situational factor data, the step further includes: Cleaning the user preference data and the situational factor data, removing duplicate data and erroneous data, and preprocessing the cleaned data; The preprocessing method includes at least one of the following: Unified data format; Desensitize sensitive information; Fill in missing values.

3. The method for performing personalized prediction on users based on the CLLM model according to claim 1 or 2, characterized in that: The step of constructing a multidimensional data set based on the collected user preference data and situational factor data includes: Creating a data mapping table based on predefined rules to associate user preference data with corresponding contextual factor data; Select data storage technology and design data table structure, import user preference data and related situational factor data, and form a multidimensional data set.

4. The method for performing personalized prediction on users based on the CLLM model according to claim 1, characterized in that: The method for performing personalized prediction on users based on the CLLM model also includes: When training the CLLM model, the multidimensional dataset is divided into training set, validation set and test set; When executing the main loop of model training, the training set is used for forward propagation and backward propagation, and the model parameters are updated; After each training cycle, the validation set is used to evaluate the generalization ability of the model, monitor the training error and validation error, and generate cross-validation results; Adjust the model's hyperparameters based on the cross-validation results; After the model training is completed, the test set is used for test acceptance to evaluate the model training results.

5. The method for performing personalized prediction on users based on the CLLM model according to claim 1, characterized in that: Before the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the step also includes: By using the data collection module, when the target user initiates a request, the contextual factor data related to the target user's behavior is automatically captured; and key features are extracted from the target user's historical behavior records to form historical data; Preprocess the situational factor data and historical data corresponding to the target users so that the processed data meets the input requirements of the CLLM model; The situational factor data and historical data corresponding to the target users are input into the trained CLLM model.

6. The method for performing personalized prediction on users based on the CLLM model according to claim 1 or 5, characterized in that: After the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the method further includes: Obtain feedback from target users regarding personalized results; Conduct semantic analysis on feedback information, mark feedback types and relevance, and generate feedback data; The feedback data is associated with the data related to the personalized result, and the CLLM model is updated based on the associated result.

7. The method for performing personalized prediction on users based on the CLLM model according to claim 1, characterized in that: After the step of using the trained CLLM model to perform data analysis on the contextual factor data related to the target user's behavior and predicting and generating personalized results that meet the target user's preferences, the method further includes: Establish a model update cycle to retrain the model regularly; and, An online learning mechanism is introduced to make instant adjustments to the model using real-time data streams.

8. A control device, characterized in that: include: A construction module, configured to construct a multidimensional data set based on the collected user preference data and situational factor data; wherein the situational factor data is related to the user activity record, the user activity record is used to generate the user preference data, and the situational factor data includes at least one of the behavior time, the behavior location and the local weather; A training module, used to train the pre-built CLLM model based on the multi-dimensional data set, so that the CLLM model learns the correlation between the user preference data and the situational factor data, and has the ability to output personalized natural language based on the correlation; The prediction module is used to use the trained CLLM model to perform data analysis on contextual factor data related to the target user's behavior, and predict and generate personalized results that meet the target user's preferences.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, the steps of the method for performing personalized prediction for users based on the CLLM model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for performing personalized prediction for users based on the CLLM model as described in any one of claims 1 to 7.

Images