Intelligent recommendation engine system based on AI algorithm
By designing an intelligent recommendation engine system based on AI algorithms, using data collection, user modeling, item modeling, feature engineering, recommendation algorithm and algorithm evaluation modules, the problem of poor performance of existing recommendation systems when selecting recommendation algorithms is solved, and more efficient recommendation effects and system performance are achieved.
Patent Information
- Application Number
- CN202411898295.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-23
- Publication Date
- 2025-05-30
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When selecting a recommendation algorithm, existing recommendation systems usually randomly select one or use multiple algorithms in combination, resulting in poor recommendation results or increase the system operation burden and reduce recommendation efficiency.
Design an intelligent recommendation engine system based on AI algorithms, including data collection module, user modeling module, item modeling module, feature engineering module, recommendation algorithm module, algorithm evaluation module and selection module. Through these modules, the system can collect and analyze the data of users and items, build a feature matrix of users and items, generate recommendation results using collaborative filtering algorithms and content filtering algorithms, and evaluate the recommendation effect through the algorithm evaluation module, and finally select the best recommendation algorithm.
The system can effectively evaluate the effectiveness of multiple recommendation algorithms and select the best algorithm to use, thereby improving the accuracy of recommendations and user satisfaction and reducing the burden of system operation.
Smart Images
Figure CN120067430A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of recommendation engine systems, and particularly relates to an intelligent recommendation engine system based on AI algorithms. Background Art
[0002] An intelligent recommendation engine system is a system that uses algorithms and data analysis techniques to screen out personalized recommended content from a large amount of information by analyzing user behavior, preferences, and historical data. Such systems are widely used in fields such as e-commerce, social media, music, and video streaming, aiming to improve the user experience, promote sales, and enhance platform stickiness;
[0003] The existing technologies have the following defects:
[0004] Existing recommendation systems usually randomly select one or combine multiple recommendation algorithms for use. When randomly selecting one recommendation algorithm for use, it may lead to poor recommendation effects for enterprises. When selecting multiple recommendation algorithms for combined use, it will increase the operating burden of the engine system and reduce the recommendation efficiency. Summary of the Invention
[0005] The purpose of the present invention is to provide an intelligent recommendation engine system based on AI algorithms to solve the deficiencies in the background art.
[0006] To achieve the above purpose, the present invention provides the following technical solution: An intelligent recommendation engine system based on AI algorithms, including a data collection module, a user modeling module, an item modeling module, a feature engineering module, a recommendation algorithm module, an algorithm evaluation module, and a selection module;
[0007] Data collection module: Collect and store user behavior data, item information, and user attribute data;
[0008] User modeling module: Build a user profile by analyzing the historical behavior and attributes of users, including user interests, preferences, and behavior pattern information;
[0009] Item modeling module: After analyzing item information, model the attributes and features of recommended items, including item tags, categories, and content information;
[0010] Feature engineering module: Convert user and item information into feature data;
[0011] Recommendation algorithm module: Generate recommendation results using user and item feature data through recommendation algorithms, and the recommendation algorithms include collaborative filtering algorithms and content filtering algorithms;
[0012] Algorithm Evaluation Module: After generating personalized recommendation results for users through collaborative filtering algorithm and content filtering algorithm, obtain the historical behavior data of the two recommendation results, and evaluate the recommendation effects of the two recommendation results by analyzing the behavior data;
[0013] Selection Module: Select either the collaborative filtering algorithm or the content filtering algorithm as the recommendation algorithm for the engine system according to the evaluation results.
[0014] In a preferred embodiment, the User Modeling Module identifies the interests of users, including the purchase behavior and browsing frequency of users, establishes a user attribute model based on the attribute data of users, identifies the behavior patterns of users, and integrates the output of the rule engine into a user profile data structure, integrating the interest model, attribute model, and behavior patterns of users into a comprehensive user profile.
[0015] In a preferred embodiment, the Item Modeling Module analyzes various information of items, including tags, categories, content descriptions, attributes, extracts features from the item information, including performing one-hot encoding on tags, converting category information into numerical representations, converting the attributes and features of items into vector forms, mapping the attributes of items to a multi-dimensional vector space, calculating the similarity between items, analyzing the content association relationship between items, and analyzing the popularity of items.
[0016] In a preferred embodiment, the steps for the Recommendation Algorithm Module to generate personalized recommendation results by using the feature data of users and items through the collaborative filtering algorithm are as follows:
[0017] Construct a user-item matrix with the feature data of users and items, where each row represents a user and each column represents an item. Calculate the similarity between users or items based on similarity metrics, which can be completed by calculating the user-user similarity matrix or the item-item similarity matrix. Predict the ratings of target users for items that have not been interacted with based on the historical behaviors of other users similar to the target user, generate recommendation results by weighting the ratings of similar users. Predict the ratings of target users for items that have not been interacted with based on the similarity between the target item and other items, generate recommendation results by weighting the ratings of similar items. Filter and sort the generated recommendation results, and select the Top-N items as the final recommendation results.
[0018] In a preferred embodiment, the steps for the Recommendation Algorithm Module to generate personalized recommendation results by using the feature data of users and items through the content filtering algorithm are as follows:
[0019] Construct the feature data of users and items into a user-item matrix, where each row represents a user and each column represents an item. Convert the feature data of users and items into vector representations. Based on the similarity of feature vectors, calculate the similarity between users or items. For a given user, predict the user's interest in items that have not been interacted with according to the user's feature vector and similar users. For a given item, predict the user's interest in items that have not been interacted with according to the item's feature vector and similar items. Filter and sort the generated recommendation results, and select the top-N items as the final recommendation results.
[0020] In a preferred embodiment, the algorithm evaluation module obtains the historical behavior data of two recommendation results, and the historical behavior data includes error rate, recall rate, conversion rate, and information gain index.
[0021] In a preferred embodiment, the algorithm evaluation module comprehensively calculates the error rate, recall rate, conversion rate, and information gain index to obtain a recommendation coefficient, and the function expression is:
[0022]
[0023] In the formula, ZQ, ZH, ZL, and XS are the error rate, recall rate, conversion rate, and information gain index respectively, n is the historical sampling point of the recommendation result, each sampling point is a time period, XSi represents the information gain index of the i-th sampling point, and α, β, γ, and δ are the proportionality coefficients of the error rate, recall rate, conversion rate, and information gain index respectively, and α, β, γ, and δ are all greater than 0.
[0024] In a preferred embodiment, after the selection module obtains the recommendation coefficient of the collaborative filtering algorithm recommendation result and the recommendation coefficient of the content filtering algorithm recommendation result;
[0025] If the recommendation coefficient of the collaborative filtering algorithm recommendation result is greater than the recommendation coefficient of the content filtering algorithm recommendation result, select the collaborative filtering algorithm recommendation result for use;
[0026] If the recommendation coefficient of the collaborative filtering algorithm recommendation result is less than the recommendation coefficient of the content filtering algorithm recommendation result, select the content filtering algorithm recommendation result for use;
[0027] If the recommendation coefficient of the collaborative filtering algorithm recommendation result is equal to the recommendation coefficient of the content filtering algorithm recommendation result, either one can be selected for use;
[0028] If the recommendation coefficients of both the collaborative filtering algorithm recommendation result and the content filtering algorithm recommendation result are less than the recommendation threshold, replace with other recommendation algorithms.
[0029] In the above technical solution, the technical effects and advantages provided by the present invention:
[0030] 1. The present invention collects and stores user behavior data, item information, and user attribute data through a data collection module. The feature engineering module converts the information of users and items into feature data. The recommendation algorithm module generates recommendation results by using the feature data of users and items through recommendation algorithms, including collaborative filtering algorithms and content filtering algorithms. After the algorithm evaluation module generates personalized recommendation results for users through collaborative filtering algorithms and content filtering algorithms, it obtains the historical behavior data of the two recommendation results, and evaluates the recommendation effects of the two recommendation results by analyzing the behavior data. The selection module selects a collaborative filtering algorithm or a content filtering algorithm as the recommendation algorithm of the engine system according to the evaluation results. This recommendation system can effectively evaluate the effects of the recommendation results generated by multiple recommendation algorithms, enabling enterprises to select the best recommendation algorithm for use. Enterprises can provide more personalized recommendations that meet the interests and preferences of users, thereby improving user satisfaction and experience.
[0031] 2. The present invention obtains the historical behavior data of the two recommendation results through the algorithm evaluation module. The historical behavior data includes error rate, recall rate, conversion rate, and information gain index. The recommendation coefficient is obtained by comprehensively calculating the error rate, recall rate, conversion rate, and information gain index. According to the evaluation results, the filtering algorithm with a larger recommendation coefficient is selected as the recommendation algorithm of the engine system for use, effectively improving the recommendation effect and recommendation accuracy of the engine system. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.
[0033] Figure 1 It is the system module diagram of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0035] Embodiment: Please refer to Figure 1As shown in the figure, the intelligent recommendation engine system based on AI algorithm described in this embodiment includes a data collection module, a user modeling module, an item modeling module, a feature engineering module, a recommendation algorithm module, an algorithm evaluation module, and a selection module;
[0036] Data collection module: This module is responsible for collecting and storing user behavior data, item information, and user attribute data. The user behavior data and user attribute data are sent to the user modeling module, the item information is sent to the item modeling module, and the user behavior data, item information, and user attribute data are sent to the feature engineering module;
[0037] Clearly define data requirements: Before designing the data collection module, it is necessary to clearly define the data required by the recommendation system. This may include user behavior data (clicks, views, purchases, etc.), item information (tags, categories, descriptions, etc.), user attribute data (age, gender, geographical location, etc.), and so on.
[0038] Select appropriate data sources: Determine the data sources, which may involve obtaining data from channels such as websites, mobile applications, and social media. Different data sources may require different access methods and protocols.
[0039] Design data collection interfaces: Develop data collection interfaces to ensure that the required data can be effectively collected from the data sources. This may involve integrating with website analysis tools, databases, APIs, etc.
[0040] User behavior data collection: During the interaction between users and the system, record key user behavior data, such as clicking on products, browsing pages, adding to the shopping cart, purchasing, etc. These data are usually used for user behavior analysis and personalized recommendations.
[0041] Item information data collection: Collect information about items, including item features, attributes, tags, etc. This information helps to construct the feature vectors of items for the training and prediction of recommendation algorithms.
[0042] User attribute data collection: Obtain user-related attribute information, such as information provided during registration and user information on social media. This helps to build user portraits and improve the personalization level of recommendations.
[0043] Real-time data stream processing: For systems that require real-time recommendations, real-time data stream processing may need to be considered to ensure that the recommendation model can adapt to new user behavior data in a timely manner.
[0044] Data cleaning and preprocessing: Clean and preprocess the collected data, including handling missing values, removing duplicates, and dealing with outliers, to ensure the quality and usability of the data.
[0045] Store data: Store the collected data in an appropriate data storage system, such as a relational database, a NoSQL database, a data warehouse, etc. The choice of the appropriate storage system depends on the scale of the data and the query requirements.
[0046] Protect user privacy: During the data collection process, ensure that the user privacy is effectively protected. Use appropriate privacy protection measures, such as data desensitization, anonymization, etc.
[0047] Build data indexes: To improve the data retrieval efficiency, indexes can be built to accelerate the query operations on user behavior, item information, and user attribute data.
[0048] Monitoring and logging: Deploy a monitoring system and a logging mechanism to track the running status of the data collection module at any time and discover and solve problems in a timely manner.
[0049] User modeling module: By analyzing the user's historical behavior and attributes, construct a user profile, including the user's interests, preferences, behavior pattern information, and send the user profile to the cloud for storage;
[0050] Rule definition: Clearly define a set of rules based on the user's historical behavior, attribute data, and behavior patterns. The rules can be logical judgments based on business knowledge, such as the user has purchased a certain type of product or accessed the platform within a specific time period.
[0051] User interest model: Design rules to identify the user's interests. This may include the user's purchase behavior under a specific category, browsing frequency, etc. The rule engine can determine which categories or products the user has a high interest in.
[0052] User attribute model: Design rules to establish a user attribute model based on the user's attribute data. For example, the user's age group, gender, geographical location, etc. can be determined through rules to construct the user's basic attribute characteristics.
[0053] Behavior pattern recognition: Formulate rules to identify the user's behavior patterns. This may include the user's shopping time preference during a day, the frequency of browsing paths, etc. The rule engine can judge the user's typical behavior based on these rules.
[0054] User profile integration: Integrate the user's interest model, attribute model, and behavior patterns into a comprehensive user profile. This can be achieved by integrating the output of the rule engine into a user profile data structure, which can include interest tags, attribute characteristics, behavior patterns, etc.
[0055] Item modeling module: After analyzing the item information, model the attributes and characteristics of the recommended items, which can include item tags, categories, content, etc. information, and send the item model to the cloud for storage;
[0056] Item Information Analysis: Analyze various information of the item, including labels, categories, content descriptions, attributes, etc. Understanding the multi-dimensional information of the item helps to better understand the characteristics of the item.
[0057] Feature Extraction: Extract meaningful features from the item information. This can include one-hot encoding of labels, converting category information into numerical representations, etc., so that subsequent modeling and recommendation algorithms can use these features.
[0058] Vectorization: Convert the attributes and features of the item into vector form. This can be achieved by using methods such as Word Embeddings or other representation learning methods to map the attributes of the item into a multi-dimensional vector space.
[0059] Similarity Calculation: Calculate the similarity between items. This can be achieved by calculating the cosine similarity, Euclidean distance, etc. between vectors, so as to find similar items during recommendation.
[0060] Content Association Analysis: Analyze the content association relationship between items and find items with strong correlation. This helps to provide more diverse and relevant recommendation results in the recommendation system.
[0061] Popularity Analysis: Analyze the popularity of items to understand which items are more popular among users. Popularity information can be used for recommending popular items in the recommendation system or adjusting the weights of personalized recommendations.
[0062] Dynamic Update of the Model: The attributes and features of items may change over time. Therefore, it is necessary to update the item model regularly to reflect the latest characteristics of the items.
[0063] Model Interpretability: Consider increasing the interpretability of the model so that the output of the item model is more understandable to business personnel, especially in scenarios where it is necessary to explain the recommendation results.
[0064] Protect Item Information: During the process of building the item model, ensure that appropriate privacy protection measures are adopted to minimize the leakage of sensitive information.
[0065] Feature Engineering Module: Convert the information of users and items into feature data that can be understood and processed, and send the feature data to the recommendation algorithm module;
[0066] Feature Extraction: Extract meaningful features from the original information of users and items. This may include extracting statistical information, frequency distributions, etc. of user and item attributes.
[0067] One-Hot Encoding: For features with discrete categories, methods such as One-Hot Encoding are used to convert them into numerical forms that the model can handle. For example, the gender of a user, the category of an item, etc.
[0068] Numericalization: Numericalize some features with sequential or size relationships so that the model can understand. For example, the age of a user, the price of an item, etc.
[0069] Time Features: If time-related information is involved, time features such as hour, day of the week, month, etc. can be extracted to better capture the impact of time.
[0070] Text Feature Processing: If the information of a user or an item contains text data, text feature processing can be performed, such as extracting keywords, using the bag-of-words model, TF-IDF, etc.
[0071] Combined Features: New combined features can be created according to business logic and the relationships between features to provide more information. For example, the product of the user's purchase frequency and the average price of an item.
[0072] Standardization / Normalization: Standardize or normalize features to ensure that different features have similar numerical ranges, which helps the training convergence and performance of some models.
[0073] Processing Multimodal Data: If the system contains multiple types of data (such as images, text, numerical values, etc.), it is necessary to consider how to process this multimodal data, and the way of fusing features can be adopted.
[0074] Feature Selection: When building a feature model, feature selection can be performed to exclude some features that are not helpful or redundant for model prediction to reduce the complexity of the model.
[0075] Dynamic Feature Update: The distribution and importance of features may change over time, so features need to be updated regularly.
[0076] Recommendation Algorithm Module: Use the feature data of users and items through recommendation algorithms to generate personalized recommendation results. Recommendation algorithms include collaborative filtering algorithms and content filtering algorithms;
[0077] Using the feature data of users and items through collaborative filtering algorithms to generate personalized recommendation results includes the following steps:
[0078] User-Item Matrix Construction: Construct a user-item matrix with the feature data of users and items, where each row represents a user and each column represents an item. The elements in the matrix can be the ratings of users for items or other relevant metrics.
[0079] Similarity calculation: Based on the selected similarity metric, calculate the similarity between users or items. This can be done by calculating the user-user similarity matrix or the item-item similarity matrix.
[0080] Recommendation generation: For a given user, use the collaborative filtering algorithm to predict the ratings or probabilities of items that the user has not interacted with. This can be done through the following two main methods:
[0081] User collaborative filtering: Predict the ratings of items that the target user has not interacted with based on the historical behavior of other users who are similar to the target user. Generate the recommendation results by weighting the ratings of similar users.
[0082] Item collaborative filtering: Predict the ratings of items that the target user has not interacted with based on the similarity between the target item and other items. Generate the recommendation results by weighting the ratings of similar items.
[0083] Filtering and sorting: Filter and sort the generated recommendation results, and select the top-N items as the final recommendation results. Sorting can be performed according to metrics such as predicted ratings and similarity.
[0084] Generating personalized recommendation results using user and item feature data through content filtering algorithms includes the following steps:
[0085] User-item matrix construction: Construct a user-item matrix from the feature data of users and items, where each row represents a user, each column represents an item, and the elements in the matrix can be the ratings of users for items or other relevant metrics.
[0086] Feature vector representation: Convert the feature data of users and items into vector representations. This can be done by combining the features of users and items to form a vector, and each element of the vector represents a feature.
[0087] Similarity calculation: Based on the similarity of feature vectors, calculate the similarity between users or items. Commonly used similarity metrics include cosine similarity, Euclidean distance, Jaccard similarity, etc.
[0088] Item similarity calculation: For item recommendation, calculate the similarity between items to find items with similar features.
[0089] User interest prediction: For a given user, predict the user's interest in items that the user has not interacted with based on the user's feature vector and similar users (if any).
[0090] Item recommendation: For a given item, predict the user's interest in items that the user has not interacted with based on the item's feature vector and similar items.
[0091] Filtering and Sorting: Filter and sort the generated recommendation results, and select the top-N items as the final recommendation results. Sorting can be performed according to indicators such as predicted interest and similarity.
[0092] Algorithm Evaluation Module: After generating personalized recommendation results for users through collaborative filtering algorithm and content filtering algorithm, obtain the historical behavior data of the two recommendation results, and evaluate the recommendation effects of the two recommendation results by analyzing the behavior data. The evaluation results are sent to the selection module and the administrator;
[0093] The algorithm evaluation module obtains the historical behavior data of the two recommendation results. The historical behavior data includes error rate, recall rate, conversion rate, and information gain index;
[0094] The algorithm evaluation module comprehensively calculates the error rate, recall rate, conversion rate, and information gain index to obtain a recommendation coefficient. The function expression is:
[0095]
[0096] In the formula, TJX is the recommendation coefficient, ZQ, ZH, ZL, and XS are the error rate, recall rate, conversion rate, and information gain index respectively, n is the historical sampling point of the recommendation result, each sampling point is a time period, XSi represents the information gain index of the i-th sampling point, and α, β, γ, and δ are the proportionality coefficients of the error rate, recall rate, conversion rate, and information gain index respectively, and α, β, γ, and δ are all greater than 0;
[0097] This application obtains the historical behavior data of the two recommendation results through the algorithm evaluation module. The historical behavior data includes error rate, recall rate, conversion rate, and information gain index. The error rate, recall rate, conversion rate, and information gain index are comprehensively calculated to obtain a recommendation coefficient. According to the evaluation results, the filtering algorithm with a larger recommendation coefficient is selected as the recommendation algorithm of the engine system, effectively improving the recommendation effect and recommendation accuracy of the engine system.
[0098] Error Rate:
[0099] Definition: The historical error rate usually refers to the degree of mismatch between the items actually interacted by users and the items in the recommendation results generated by the recommendation system in the past period.
[0100] Obtaining Method: Collect historical recommendation results and actual behavior data of users, compare whether the recommended items match the items actually interacted by users, and calculate and obtain the error rate.
[0101] Recall Rate:
[0102] Definition: The historical recall rate refers to the proportion of the items of interest to users successfully found by the recommendation system in the past period to all the items of interest to users actually.
[0103] Obtaining method: For each user, collect the items that have been actually interacted with, then compare whether these items appear in the recommended results, and calculate the historical recall rate.
[0104] Conversion rate:
[0105] Definition: The historical conversion rate refers to the proportion of the recommended clicks or interaction behaviors generated by the user through the recommendation system that are finally converted into actual purchase or other target behaviors.
[0106] Obtaining method: Collect the behaviors of the user such as clicks, interactions, and purchases in the recommendation system, count the conversion situations of these behaviors, and calculate the historical conversion rate.
[0107] Information gain index:
[0108] Definition: The information gain index measures whether the recommended results of the recommendation system provide new information about the user's behavior.
[0109] Obtaining method: An A / B test or experimental design is required. Compare the effect of the recommendation system with a baseline model or a control group, and then calculate the information gain index by comparing the effects of the recommendation system and the control group.
[0110] Selection module: Select the collaborative filtering algorithm or the content filtering algorithm as the recommendation algorithm of the engine system according to the evaluation results;
[0111] After obtaining the recommendation coefficient of the collaborative filtering algorithm recommendation result and the recommendation coefficient of the content filtering algorithm recommendation result;
[0112] If the recommendation coefficient of the collaborative filtering algorithm recommendation result is greater than the recommendation coefficient of the content filtering algorithm recommendation result, select the collaborative filtering algorithm recommendation result for use;
[0113] If the recommendation coefficient of the collaborative filtering algorithm recommendation result is less than the recommendation coefficient of the content filtering algorithm recommendation result, select the content filtering algorithm recommendation result for use;
[0114] If the recommendation coefficient of the collaborative filtering algorithm recommendation result is equal to the recommendation coefficient of the content filtering algorithm recommendation result, either one can be selected for use;
[0115] If the recommendation coefficient of the collaborative filtering algorithm recommendation result and the recommendation coefficient of the content filtering algorithm recommendation result are both less than the recommendation threshold, replace with other recommendation algorithms such as machine learning and neural networks.
[0116] This application collects and stores user behavior data, item information, and user attribute data through a data collection module. The feature engineering module converts the information of users and items into feature data. The recommendation algorithm module uses the feature data of users and items through a recommendation algorithm to generate recommendation results. The recommendation algorithms include a collaborative filtering algorithm and a content filtering algorithm. After the algorithm evaluation module generates personalized recommendation results for users through the collaborative filtering algorithm and the content filtering algorithm, it obtains the historical behavior data of the two recommendation results, and evaluates the recommendation effects of the two recommendation results by analyzing the behavior data. The selection module selects the collaborative filtering algorithm or the content filtering algorithm as the recommendation algorithm of the engine system according to the evaluation results. This recommendation system can effectively evaluate the effects of the recommendation results generated by multiple recommendation algorithms, enabling enterprises to select the best recommendation algorithm for use. Enterprises can provide more personalized recommendations that match users' interests and preferences, thereby improving user satisfaction and experience.
[0117] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.
[0118] In the description of this specification, the description referring to terms such as "one embodiment", "example", "specific example", etc. means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples.
[0119] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not elaborate on all details and do not limit the present invention to only the specific implementation manners. Obviously, many modifications and variations can be made according to the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the present invention, so that those skilled in the art in the relevant technical field can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.
Claims
1. An intelligent recommendation engine system based on AI algorithm, characterized by: It includes data collection module, user modeling module, item modeling module, feature engineering module, recommendation algorithm module, algorithm evaluation module and selection module; Data collection module: collects and stores user behavior data, item information, and user attribute data; User modeling module: builds user portraits by analyzing the user's historical behavior and attributes, including user interests, preferences, and behavior pattern information; Item modeling module: After analyzing item information, it models the attributes and features of recommended items, including item labels, categories, and content information; Feature engineering module: converts user and item information into feature data; Recommendation algorithm module: Generates recommendation results using the feature data of users and items through recommendation algorithms, including collaborative filtering algorithms and content filtering algorithms; Algorithm evaluation module: After generating personalized recommendation results for users through collaborative filtering algorithm and content filtering algorithm, the historical behavior data of the two recommendation results are obtained, and the recommendation effects of the two recommendation results are evaluated by analyzing the behavior data; Selection module: Select collaborative filtering algorithm or content filtering algorithm as the recommendation algorithm of the engine system based on the evaluation results.
2. The intelligent recommendation engine system based on AI algorithm according to claim 1, characterized in that: The user modeling module identifies the user's interests, including the user's purchasing behavior and browsing frequency, establishes a user attribute model based on the user's attribute data, identifies the user's behavior pattern, and integrates the user's interest model, attribute model and behavior pattern into a comprehensive user portrait by integrating the output of the rule engine into a user portrait data structure.
3. The intelligent recommendation engine system based on AI algorithm according to claim 2, characterized in that: The item modeling module analyzes various information of the item, including tags, categories, content descriptions, and attributes, and extracts features from the item information, including unique-hot encoding of tags, converting category information into numerical representations, converting the attributes and features of the item into vector form, mapping the attributes of the item to a multidimensional vector space, calculating the similarity between items, analyzing the content association relationship between items, and analyzing the popularity of items.
4. The intelligent recommendation engine system based on AI algorithm according to claim 1, characterized in that: The recommendation algorithm module generates personalized recommendation results by using the feature data of users and items through a collaborative filtering algorithm, including the following steps: The feature data of users and items are constructed into a user-item matrix, in which each row represents a user and each column represents an item. The similarity between users or items is calculated based on the similarity metric, which can be accomplished by calculating the user-user similarity matrix or the item-item similarity matrix. The target user's score for uninteracted items is predicted based on the historical behavior of other users similar to the target user. The recommendation results are generated by weighting the scores of similar users. The target user's score for uninteracted items is predicted based on the similarity between the target item and other items. The recommendation results are generated by weighting the scores of similar items. The generated recommendation results are filtered and sorted, and the Top-N items are selected as the final recommendation results.
5. The intelligent recommendation engine system based on AI algorithm according to claim 4, characterized in that: The recommendation algorithm module generates personalized recommendation results by using the feature data of users and items through a content filtering algorithm, including the following steps: The feature data of users and items are constructed into a user-item matrix, where each row represents a user and each column represents an item. The feature data of users and items are converted into vector representations. Based on the similarity of feature vectors, the similarity between users or items is calculated. For a given user, the user's interest in items with which they have not interacted is predicted based on the user's feature vector and similar users. For a given item, the user's interest in items with which they have not interacted is predicted based on the item's feature vector and similar items. The generated recommendation results are filtered and sorted, and the Top-N items are selected as the final recommendation results.
6. The intelligent recommendation engine system based on AI algorithm according to claim 5, characterized in that: The algorithm evaluation module obtains historical behavior data of two recommendation results, and the historical behavior data includes error rate, recall rate, conversion rate and information gain index.
7. The intelligent recommendation engine system based on AI algorithm according to claim 6, characterized in that: The algorithm evaluation module calculates the error rate, recall rate, conversion rate and information gain index to obtain the recommendation coefficient. The function expression is: Where ZQ, ZH, ZL, and XS are the error rate, recall rate, conversion rate, and information gain index, respectively; n is the historical sampling point of the recommendation result, each sampling point is a time period; XSi represents the information gain index of the i-th sampling point; α, β, γ, and δ are the proportional coefficients of the error rate, recall rate, conversion rate, and information gain index, respectively; and α, β, γ, and δ are all greater than 0.
8. The intelligent recommendation engine system based on AI algorithm according to claim 7, characterized in that: After the selection module obtains the recommendation coefficient of the collaborative filtering algorithm recommendation result and the recommendation coefficient of the content filtering algorithm recommendation result; If the recommendation coefficient of the collaborative filtering algorithm recommendation result is greater than the recommendation coefficient of the content filtering algorithm recommendation result, the collaborative filtering algorithm recommendation result is selected for use; If the recommendation coefficient of the collaborative filtering algorithm recommendation result is smaller than that of the content filtering algorithm recommendation result, the content filtering algorithm recommendation result is selected for use; If the recommendation coefficient of the collaborative filtering algorithm recommendation result is equal to the recommendation coefficient of the content filtering algorithm recommendation result, then select one of them for use; If the recommendation coefficient of the collaborative filtering algorithm recommendation result and the recommendation coefficient of the content filtering algorithm recommendation result are both less than the recommendation threshold, then other recommendation algorithms are replaced.