Commodity recommendation method and system based on big data analysis

Abstract

The invention discloses a commodity recommendation method and system based on big data analysis, and relates to the technical field of intelligent recommendation systems.The method comprises the steps that behavior data of a user on a platform and store data of a region where the user is located are collected; public opinion data of the commodities are collected from the social platform and the comment area, and the behavior data, the store data and the public opinion data are integrated to form a recommendation data set; extracting behavior features from the behavior data, extracting store features from the store data, and extracting public opinion features from the public opinion data; integrating the behavior features, the store features and the public opinion features into a comprehensive feature vector, and generating a user portrait according to the comprehensive feature vector; based on the user portrait, utilizing a machine learning algorithm to construct a commodity recommendation model, taking the user portrait as input, and outputting a commodity recommendation list; and behavior data and real-time data of store activities are collected, a recommendation list is dynamically adjusted, and feedback of the user to recommended commodities is recorded.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent recommendation systems, and particularly to a commodity recommendation method and system based on big data analysis. Background Art

[0002] With the rapid development of e-commerce and online retail platforms, recommendation systems have become one of the key technologies to improve user experience and commodity sales. Traditional recommendation methods such as collaborative filtering and content-based recommendation algorithms can achieve the recommendation function to a certain extent, but in the face of large-scale and multi-dimensional data, there are problems of insufficient accuracy and poor flexibility. In recent years, with the help of big data and artificial intelligence technologies, recommendation systems have gradually begun to integrate user behavior data, commodity information, social public opinion data and location-related data, and try to depict user portraits through multi-dimensional data to achieve more personalized recommendations. However, these technologies still face the bottlenecks of the complexity of multi-source data integration, insufficient real-time processing efficiency and slow response to the dynamic changes of user interests in practical applications.

[0003] Although existing recommendation technologies have made certain progress in improving recommendation quality, there are still obvious deficiencies. The analysis of user behavior data is mostly limited to simple click or purchase records, lacking in-depth exploration of the temporal characteristics of behavior and dynamic interests; the utilization of store data more depends on static geographical location relevance, and fails to fully combine the dynamic influence of store activities within the region; for the processing of public opinion data, the comprehensive effects of emotional tendency intensity and public opinion heat are often ignored. In addition, existing technologies are relatively single in the integration of multi-source heterogeneous data and the mining of interaction relationships, resulting in insufficient recommendation accuracy and personalization. At the same time, existing methods generally lack the ability of real-time update, and it is difficult to dynamically adjust recommendation strategies according to changes in user behavior or store activities, and it is difficult to meet the high demand of users for accurate recommendations. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] The present invention provides a commodity recommendation method based on big data analysis to solve the problem that existing technologies are relatively single in the integration of multi-source heterogeneous data and the mining of interaction relationships, resulting in insufficient recommendation accuracy and personalization.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a commodity recommendation method based on big data analysis, which includes: Collecting the behavior data of users on the platform and the store data of the regions where the users are located; collecting the public opinion data of commodities from social platforms and comment areas, integrating the behavior data, store data and public opinion data to form a recommendation data set and performing preprocessing; Extract behavioral features from behavioral data, store features from store data, and extract public opinion features from public opinion data; Integrate behavioral features, store features, and public opinion features into a comprehensive feature vector, and generate a user profile based on the comprehensive feature vector; Based on the user profile, use machine learning algorithms to build a product recommendation model. Use the user profile as the input of the model, and combine the public opinion popularity of the product and the store activity data in the region. The product recommendation model outputs a recommended list of products; Collect real-time data of behavioral data and store activities, dynamically adjust the recommended list, and record the user's feedback on the recommended products.

[0007] As a preferred solution of the product recommendation method based on big data analysis according to the present invention, wherein: collect the user's behavioral data on the platform and the store data of the user's location; collect the public opinion data of the product from social platforms and comment areas. The specific steps are as follows: Combine the IP address of the user device and the GPS positioning data to initially obtain the user's geographical location; Bind the user identifier and the geographical location, and initially establish a record; Obtain the user's behavioral data on the platform through the product recommendation platform; According to the user's geographical location, screen the stores in the user's area from the store database; Collect public opinion data related to the product from social platforms and comment areas.

[0008] As a preferred solution of the product recommendation method based on big data analysis according to the present invention, wherein: integrate the behavioral data, store data, and public opinion data to form a recommendation data set and perform preprocessing. Specifically: Integrate the behavioral data, store data, and public opinion data to form a recommendation data set. The expression is: ; Wherein, Represents the recommendation data set of the th user, Represents the behavioral data of the th user, Represents the store data related to the th user, Represents the public opinion data related to the th user; Use a hash table to generate a unique hash value for each piece of data in the recommendation data set, compare whether the hash values are repeated, and delete the repeated data after hash conflict detection; Detect outliers in the recommended dataset using the interquartile range (IQR), correct the outliers, set the outlier threshold, and correct the data exceeding the outlier threshold using the mean of similar data; Predict the values of missing data based on the values of the k nearest neighbor points most similar to the current data point through the k-nearest neighbor algorithm, and then complete the missing data through KNN interpolation.

[0009] As a preferred solution of the commodity recommendation method based on big data analysis described in the present invention, wherein: extracting behavioral features from behavioral data, extracting store features from store data, and extracting public opinion features from public opinion data, the specific steps are as follows: Perform aggregation analysis on the behavioral data, count the behavioral frequencies of users within a certain time window, and extract the distribution characteristics of behavioral types. The behavioral feature extraction formula is: ; Wherein, represents the feature vector of the behavioral data, represents the length of the time window, , , and respectively represent the number of click, search, browse, and purchase behaviors of the user at time , represents the user behavior time period feature; Statistically analyze the number of product listings and classification distributions of stores in the user's area, and extract the matching characteristics of product categories and inventory. The store feature extraction formula is: ; Wherein, represents the feature vector of the store data, represents the number of stores in the user's area, represents the index variable of the store, represents the th number of on-sale products in the store, represents the th activity discount intensity of the store, represents the straight-line distance between the user and the th store, represents the distance weight function; Statistically analyze the number of mentions of products on social media, and extract the popularity features of products; analyze the comment content through a sentiment analysis model, and extract the sentiment tendency scores of products; combine the product popularity and sentiment tendency scores to calculate the evaluation features of products. The public opinion feature extraction formula is: ; Wherein, Represents the feature vector of public opinion data, Represents the total number of public opinion data statistically, Represents the Forwarding volume of the Represents the th public opinion data; The integration of behavioral features, store features, and public opinion features into a comprehensive feature vector, and generating a user portrait based on the comprehensive feature vector. The specific steps are as follows: The feature vector integration formula is: ; Among them, Represents the integrated comprehensive feature vector; Based on the comprehensive feature vector Construct a user portrait for each user.

[0010] As a preferred solution of the commodity recommendation method based on big data analysis described in the present invention, wherein: based on the user portrait, using a machine learning algorithm to construct a commodity recommendation model. The specific steps are as follows: Adopt the DeepFM model based on deep learning as the commodity recommendation model, combine the user portrait, commodity features, and dynamic features, and output the predicted interest score of the user for the commodity; The commodity features mainly refer to the basic information of the commodity and the public opinion heat feature. The public opinion heat feature calculation formula is: ; Among them, Represents the public opinion heat score of the commodity, Represents the index variable of the public opinion record, Represents the th forwarding volume of the public opinion data, Represents the th emotional tendency score of the public opinion data, Represents the logarithmic function, Represents the th comment number of the public opinion data; The dynamic features refer to the real-time update situation of the store activity data and behavior data in the region. The regional store activity feature can be calculated by the following formula: ; Among them, Represents the store activity intensity in the region.

[0011] As a preferred solution of the commodity recommendation method based on big data analysis of the present invention, wherein: the user portrait is used as the input of the model, and combined with the public opinion heat of the commodity and the store activity data in the region, the commodity recommendation model outputs a recommended list of commodities. The specific steps are as follows: Map the commodity features and dynamic features to a low-dimensional dense vector space to generate feature embedding representations; For the feature cross FM part of the commodity recommendation model, use a factorization machine to capture the second-order interaction relationship between the input features to obtain the output of the FM part ; For the non-linear feature learning Deep part of the commodity recommendation model, use a fully connected deep neural network to extract high-order non-linear feature interactions to obtain the output of the Deep part ; Perform weighted fusion on the outputs of the FM part and the Deep part to obtain the final interest prediction score , and the expression is: ; Wherein, represents the Sigmoid function; According to the interest prediction score , sort the interest prediction scores of all commodities in descending order, and select the highest scoring commodities to generate a recommended list , and display the recommended list to the user.

[0012] As a preferred solution of the commodity recommendation method based on big data analysis of the present invention, wherein: collect the real-time data of the behavior data and store activities, dynamically adjust the recommended list, and record the feedback of the user on the recommended commodities. The specific steps are as follows: Collect the behavior data and feedback data of the user in real time, and clean and preprocess them; Dynamically calculate the recommended priority score of each commodity according to the behavior data and feedback data of the user , and the expression is: ; Wherein, represents the total number of times the user clicks on the commodity , represents the total number of times the user purchases the commodity , represents the total stay time of the user on the page of the commodity , represents the score of the user on the commodity , represents the base of the natural logarithm, and respectively represent the feature vectors of store data and public opinion data, representing the user behavior time period feature.

[0013] In a second aspect, the present invention provides a product recommendation system based on big data analysis, including an acquisition module, a feature extraction module, a product recommendation module, an adjustment feedback module, and a display module: The acquisition module is responsible for acquiring user-related data, preprocessing and integrating the data, and generating a recommendation data set; The feature extraction module is responsible for extracting key features from the acquired data, generating a user profile, and providing an input for the recommendation model; The product recommendation module constructs a recommendation model based on the user profile and product features, predicts the interest score of the user for the product, and generates a recommendation list; The adjustment feedback module is used to collect user behavior data and store activity data in real time, dynamically adjust the recommendation list, record and analyze user feedback; The display module is used to display the recommendation result to the user in a user-friendly manner and provide an interaction function.

[0014] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the product recommendation method based on big data analysis as described in the first aspect of the present invention is implemented.

[0015] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the product recommendation method based on big data analysis as described in the first aspect of the present invention is implemented.

[0016] The beneficial effects of the present invention are as follows: Through the processes of data collection, preprocessing, feature extraction, model construction, and dynamic adjustment, the present invention proposes a commodity recommendation method based on big data analysis. First, by means of the login module, user behavior data, regional store data, and commodity public opinion data are collected to construct a multi-dimensional recommendation data set, providing comprehensive data support for recommendations. Secondly, through data cleaning, duplicate removal, anomaly correction, and missing value filling, the accuracy and integrity of the data are ensured. Subsequently, user behavior characteristics, store activity characteristics, and public opinion characteristics are extracted to generate a comprehensive feature vector for constructing a user profile. Based on the user profile, a deep learning model is used to capture feature interactions, and a recommendation list is generated in combination with the dynamic characteristics of commodities. By real-time collecting user feedback and dynamically adjusting the recommendation priority, the recommendation results are further optimized. This method effectively solves the problems of poor data quality, response lag, and insufficient personalization in the prior art, significantly improving the accuracy, real-time performance, and user satisfaction of recommendations, and having important value in the field of intelligent recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings according to these drawings without creative efforts.

[0018] Figure 1 It is a flowchart of the commodity recommendation method based on big data analysis in Embodiment 1.

[0019] Figure 2 It is a schematic diagram of generating a recommendation list in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings of the specification.

[0021] In the following description, many specific details are set forth to fully understand the present invention, but the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar promotions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0022] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that mutually excludes other embodiments.

[0023] Example 1, refer to Figure 1 and Figure 2 , which is the first embodiment of the present invention. This embodiment provides a commodity recommendation method based on big data analysis, including the following steps: S1. Collect the behavior data of users on the platform and the store data of the regions where the users are located; collect the public opinion data of commodities from social platforms and comment areas, integrate the behavior data, store data and public opinion data, and form a recommendation data set for preprocessing; Combined with the IP address and GPS positioning data of the user device, initially obtain the geographical location of the user; Bind the user identification and geographical location, and initially establish a record; According to the user's geographical location, screen the stores in the user's area from the store database, including store geographical location, stocked commodities and activity information; Obtain the behavior data of the user on the platform through the commodity recommendation platform, including the number of clicks, number of views, number of searches and number of purchases, and each behavior is accompanied by a timestamp and a behavior type; The behavior types include click, view, search and purchase; According to the user's geographical location, screen the stores in the user's area from the store database, including store geographical location, stocked commodities and activity information; Collect the public opinion data related to commodities from social platforms and comment areas, including user evaluations, commodity popularity and sentiment inclination; Preprocessing means deleting duplicate data, correcting abnormal data and filling in missing data; Integrate the behavior data, store data and public opinion data to form a recommendation data set, and the expression is: ; Wherein, represents the recommendation data set of the th user, represents the behavior data of the th user, represents the store data related to the th user, represents the public opinion data related to the th user; generate a unique hash value for each piece of data in the recommendation data set using a hash table, compare whether the hash values are repeated, and delete the repeated data after hash conflict detection, so as to achieve fast and efficient deduplication; Delete the repeated records in the behavior data, store data and public opinion data to ensure the uniqueness and accuracy of the data; Use the interquartile range (IQR) to detect outliers in the recommended dataset, correct the outliers, set the outlier threshold, and correct the data exceeding the outlier threshold with the mean of similar data; The threshold for outlier correction is set to IQR × 1.5, and the expression is: ; Where, represents the third quartile of the dataset, represents the first quartile of the dataset; For example, in the user stay time data, if = 60 seconds, = 300 seconds, then the data exceeding 300 + 1.5×(300 - 60) = 660 seconds is determined as an outlier and corrected with the mean of similar users; Further explanation, IQR is an index used in statistics to measure the degree of data dispersion, defined as the difference between the third quartile ( ) and the first quartile ( ), that is: ; IQR × 1.5 is not directly used as the outlier threshold, but 1.5 times of IQR extends upward from to form a "tolerance range", and the data points exceeding this range are determined as outliers, that is, the outlier threshold is the sum of and 1.5 times of IQR.

[0024] Through the K-nearest neighbor algorithm, predict the value of the missing data according to the values of the k nearest neighbor points most similar to the current data point, and then complete the missing data through the KNN interpolation method; In the K-nearest neighbor algorithm for missing value completion, the value of k is set to 5 - 15. For example, through cross-validation, it is found that when k = 10, the interpolation accuracy (F1-score = 0.92) is the highest and the calculation efficiency is the best; The present invention constructs a multi-dimensional recommended dataset through the steps of data integration and preprocessing, behavior feature extraction, and public opinion data introduction, and combines specific feature extraction methods and algorithm optimization, significantly improving the accuracy and timeliness of recommendations. The design of each step is based on the core requirements of recommendations. At the same time, by introducing techniques such as hash table deduplication, IQR outlier correction, and KNN interpolation, the present invention is superior to the prior art in terms of data processing efficiency and recommendation effect. This technical solution not only solves the problem of user personalized recommendation, but also provides an innovative idea for the dynamic adjustment of recommendations.

[0025] S2. Extract behavior features from behavior data, extract store features from store data, and extract public opinion features from public opinion data; Conduct aggregate analysis on behavioral data, count the behavioral frequencies of users within a certain time window, extract the distribution characteristics of behavioral types, calculate the interest weights of users for different categories of goods, extract the preference characteristics of specific commodity categories, and extract the time characteristics of user behavior. The behavioral feature extraction formula is: ; Among them, represents the feature vector of behavioral data, represents the length of the time window, , , and respectively represent the number of click, search, browse, and purchase behaviors of the user at time , represents the time period feature of user behavior, with a value range of [0, 1], indicating the distribution of user behavior activity within a day; For example, if user C has click, search, browse, and purchase counts of [45, 8, 30, 2] within a 7-day time window, and the time activity h(t) = 0.82 (proportion of night-time behavior), then: ; Average daily click behavior frequency = ; Average daily search behavior frequency = ; Average daily browse behavior frequency = ; Average daily purchase behavior frequency = ; represents the distribution of user behavior activity within a day, calculated by counting the behavior frequencies of users in specific time periods (such as early morning, morning, afternoon, night).

[0026] ; In the aggregate analysis of behavioral data, the value range of the time window length is 7 to 30 days. For example, when = 7 days, it can capture the recent high-frequency behaviors of users (such as click volume during promotions); when T = 30 days, it can cover the medium- and long-term interest distributions of users (such as seasonal commodity preferences); By independently counting click, search, browse, and purchase behaviors, the interest intensity of users can be comprehensively reflected. For example, if a user frequently searches for a certain category of goods but does not purchase them, it indicates that this category of goods is a potential demand; Statistically analyze the number of product listings and classification distribution of stores within the user's region, extract the matching characteristics between product categories and inventory, extract the activity characteristics of stores (such as promotional discounts and time-limited offers), and calculate the activity intensity. Extract the distance characteristics between the geographical location of the store and the user, and measure the relevance between the store and the user through distance weights. The store feature extraction formula is: ; Among them, represents the feature vector of store data, represents the number of stores in the user's region, represents the index variable of the store, represents the th store's number of on-sale products, represents the th store's activity discount strength, with a value range of [0, 1]. The higher the value, the greater the attraction of the store's activity, represents the straight-line distance between the user and the th store, represents the distance weight function, which is used to calculate the weight of the distance on the recommendation. The value range is (0, 1]. The closer the distance, the higher the weight; combined with the user's geographical location, filter the product and activity data of nearby stores to avoid recommending products beyond the user's accessible range. For example, products in stores with a relatively close distance and a greater activity strength are more attractive, improving the relevance of the recommendation; Statistically analyze the number of times a product is mentioned on social media, and extract the popularity characteristics of the product; analyze the comment content through a sentiment analysis model, and extract the sentiment tendency score of the product; combine the product popularity and sentiment tendency score to calculate the evaluation characteristics of the product. The public opinion feature extraction formula is: ; Among them, represents the feature vector of public opinion data, represents the total number of public opinion data statistics, represents the th public opinion data's repost volume, represents the th public opinion data's sentiment tendency score (calculated through a sentiment analysis model, with a range of [0, 1]); For example, there are 100 comments. Among the 100 comments, the total number of public opinion data = 100, the average repost volume = 350, the sentiment score = 0.85, ; Total repost volume ; Assume that the repost volume of each piece of public opinion is the same: 。

[0027] ; Total sentiment score ; Assume that the sentiment scores of each piece of public opinion are the same: ; ; Since = 350 and = 0.85: ; ; ; The public opinion feature combines the popularity and sentiment evaluation of the product, enabling the recommendation to give priority to promoting products with high popularity and high evaluation. For example, a product with high popularity and good evaluation on social media is more likely to attract users to click; Through the extraction of behavioral features, store features, and public opinion features, the present invention constructs a multi-dimensional feature vector model. The design of each feature vector starts from the user's behavior, regional factors, and market feedback, solving the problems of insufficient personalization, lack of regionality, and neglect of public opinion in traditional technologies. The time sensitivity feature, distance weight function, and public opinion cross feature are innovatively introduced, significantly improving the accuracy of recommendation and the user experience.

[0028] S3. Integrate the behavioral features, store features, and public opinion features into a comprehensive feature vector, and generate a user portrait according to the comprehensive feature vector; The feature vector integration formula is: ; Among them, represents the integrated comprehensive feature vector; Use PCA (Principal Component Analysis) to compress from high dimension (such as 200 dimensions) to 128 dimensions, retaining 95% of the original information volume and improving the calculation efficiency; Calculate the covariance matrix according to the comprehensive feature vector, and the expression is: ; Among them, represents the covariance matrix, represents the total number of samples, represents the index of the sample, represents the th eigenvalue of the sample, represents the mean vector of the data set; Perform eigenvalue decomposition, with the expression: ; where represents the eigenvector matrix, represents the diagonal matrix, with dimension , and the diagonal elements are the eigenvalues arranged in descending order , and the off-diagonal elements are 0, represents the transpose of the eigenvector matrix; Select the eigenvectors (principal components) corresponding to the first largest eigenvalues, project the data into a low-dimensional space, retain the main information, and make the cumulative variance contribution rate ≥ 95%. The expression is: ; where is the th eigenvalue, is the original dimension. For example, when the original dimension is 200 and the first 128 principal components are selected, the cumulative variance accounts for 95%, and the inference speed is increased by 40%; Construct a user profile for each user based on the comprehensive eigenvector ; The generation of the comprehensive eigenvector effectively combines user behavior, geographical area restrictions, and market feedback. For example, when a user frequently browses a certain type of product (behavioral feature), and the nearby store is promoting this product (store feature), and this product has a high evaluation on social media (public opinion feature), this product can be preferentially recommended, significantly improving the hit rate and user satisfaction of the recommendation; Through integrating the behavioral feature , the store feature and the public opinion feature into the comprehensive eigenvector , the present invention constructs a multi-dimensional user profile generation method. The design of the comprehensive eigenvector solves the problems of insufficient information dimension and strong data isolation in traditional recommendation methods, significantly improving the comprehensiveness and dynamics of the user profile. In addition, by incorporating geographical location and market feedback into the user profile generation process, the performance in regional recommendation and social recommendation is optimized. Finally, the present invention can generate a multi-dimensional and high-precision user profile in a shorter time, providing a more accurate and real-time means of expressing user needs for the recommendation method.

[0029] S4. Based on the user profile, use a machine learning algorithm to construct a product recommendation model. Take the user profile as the input of the model, and combine the public opinion popularity of the product and the store activity data in the region. The product recommendation model outputs a recommendation list of products; The DeepFM model based on deep learning is adopted as the product recommendation model, which combines user portraits, product features and dynamic features to output the predicted interest score of users for products. The product features mainly refer to the basic information of the product (category, price and inventory) and the public opinion heat feature. The calculation formula of the public opinion heat feature is: ; Among them, represents the public opinion heat score of the product, represents the index variable of the public opinion record, represents the th retransmission volume of the public opinion data, represents the th sentiment tendency score of the public opinion data, represents the logarithmic function, represents the th comment number of the public opinion data; The public opinion heat threshold is set as H≥0.65. For example, when H≥0.65, the click-through rate (CTR) of the recommended product reaches 12.4%, which is 143% higher than that of the low-heat product (CTR = 5.1%); CTR is calculated by statistically counting the click behavior of users on the recommended products. The display times are the total number of times the recommended list is exposed, and the click times are the number of actual click behaviors. For example, if a product is displayed 1000 times and clicked 124 times, then the CTR is 12.4%, and the expression is: ; The dynamic features refer to the real-time update of the store activity data and behavior data in the region. The regional store activity features can be calculated by the following formula: ; Among them, represents the store activity intensity in the region; The store distance weight function adopts the exponential decay model: ; For example, when = 5km, ; = 10km, , effectively reflecting the rule that "the closer the distance, the higher the recommendation priority"; The size of depends on the combined effect of the store activity intensity and the distance weight. When the store activity intensity is high and the distance is close, the value increases; The product features and dynamic features are mapped to a low-dimensional dense vector space to generate feature embedding representations. For the FM part of the feature crossing in the product recommendation model, the factorization machine is used to capture the second-order interaction relationship between the input features, and the output of the FM part is obtained. ; For the Deep part of the non-linear feature learning in the product recommendation model, the fully connected deep neural network is used to extract the high-order non-linear feature interaction, and the output of the Deep part is obtained. ; The outputs of the FM part and the Deep part are weighted and fused to obtain the final interest prediction score. , and the expression is: ; Among them, represents the Sigmoid function, which is used to normalize the output to the interval [0, 1]; According to the interest prediction score , the interest prediction scores of all products are sorted in descending order according to , and the products with the highest scores are selected to generate a recommendation list , and the recommendation list ; Among them, represents the recommendation list, represents the number of recommended products, represents the threshold of the interest score (adjustable parameter, default value is 0.5), represents the th product, represents the user, represents the user 's interest prediction score for the product .

[0030] S5. Collect the real-time data of the behavior data and store activities, dynamically adjust the recommendation list, and record the feedback of the user on the recommended products; Collect the real-time behavior data and feedback data of the user in real time, and clean and preprocess them. The preprocessing includes deduplication and outlier processing; According to the behavior data and feedback data of the user, dynamically calculate the recommendation priority score of each product , and adjust the recommendation sequence based on the recommendation priority score. The recommendation priority score expression is: ; Among them, represents the total number of times the user clicks on the product , Indicates the total number of times a user purchases a product The total number of times, Indicates the total time a user stays on the product page, Indicates the rating of the product by the user The rating ranges from [0, 1], and the higher the value, the higher the user's satisfaction with the product. Indicates the base of the natural logarithm, and respectively represent the feature vector of store data and the feature vector of public opinion data. Indicates the user behavior time period feature; The weight distribution of the user behavior time period feature is as follows: 8:00 - 12:00, weight 0.3 (for example, the user's browsing behavior is more scattered); 12:00 - 18:00, weight 0.5 (for example, the activity level increases during lunchtime); 18:00 - 22:00, weight 0.8 (for example, the click - through rate in the evening is 3.2 times that of the early morning period); 22:00 - 8:00, weight 0.1 (for example, the user's browsing behavior is more scattered); For example, using the DeepFM model, taking the comprehensive features of user C as an example: FM part: Calculate the second - order cross - feature between product F and user C, with a weight of 0.32; The expression of the second - order cross - feature is: ; Among them, is the second - order cross - feature, and are the feature indices, is the feature dimension, , is the latent vector, and are the input feature values, is the dot product of the latent vectors.

[0031] Deep part: Extract the non - linear relationship through a 3 - layer neural network (128 → 64 → 32), and the output value is 0.68.

[0032] Final score: (Exceeding the threshold of 0.5, added to the recommendation list); In the dynamic adjustment stage, the formula for calculating the recommendation priority score of product G where: , , , , , , , , , then ; The recommended priority score of product G ≈285.44, indicating that product G will be preferentially recommended in dynamic adjustment due to high public opinion heat, strong activity intensity and positive user feedback.

[0033] The present invention dynamically calculates the recommended priority score by real-time collecting user behavior data and feedback data , and combines behavior characteristics, store characteristics and public opinion characteristics to realize the dynamic optimization of the recommendation method. The formula design of the recommended priority score comprehensively considers the depth of user interest (clicks, purchases, stay time), satisfaction (ratings), regionality (store characteristics) and market feedback (public opinion characteristics), and improves the timeliness and accuracy of the recommendation result through time sensitivity adjustment. This solution not only solves the limitations of static and single-dimensional modeling of traditional recommendation methods, but also improves the adaptability and user satisfaction of the recommendation method through the closed-loop optimization of user feedback.

[0034] This embodiment also provides a product recommendation system based on big data analysis, including: a collection module, a feature extraction module, a product recommendation module, an adjustment feedback module and a display module: The collection module is responsible for collecting user-related data, preprocessing and integrating the data, and generating a recommendation data set; The feature extraction module is responsible for extracting key features from the collected data and generating a user profile to provide input for the recommendation model; The product recommendation module constructs a recommendation model based on the user profile and product characteristics, predicts the user's interest score for the product and generates a recommendation list; The adjustment feedback module is used to real-time collect user behavior data and store activity data, dynamically adjust the recommendation list, record and analyze user feedback; The display module is used to display the recommendation result to the user in a user-friendly manner and provide an interactive function.

[0035] This embodiment also provides a computer device applicable to the situation of the product recommendation method based on big data analysis, including: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions to implement the product recommendation method based on big data analysis as proposed in the above embodiment.

[0036] The computer device may be a terminal, which includes a processor, a memory, a communication interface, a display screen, and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, carrier network, NFC (Near Field Communication), or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, trackball, or touchpad provided on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0037] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for recommending products based on big data analysis proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM for short), Electrically Erasable Programmable Read-Only Memory (EEPROM for short), Erasable Programmable Read Only Memory (EPROM for short), Programmable Read-Only Memory (PROM for short), Read-Only Memory (ROM for short), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0038] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

Claims

1. A commodity recommendation method based on big data analysis, characterized in that: include: Collect user behavior data on the platform and store data in the user's area; Collect product public opinion data from social platforms and comment areas, integrate behavior data, store data and public opinion data to form a recommendation data set and perform preprocessing; Extract behavioral features from behavioral data, extract store features from store data, and extract public opinion features from public opinion data; Integrate behavioral features, store features, and public opinion features into a comprehensive feature vector, and generate a user profile based on the comprehensive feature vector; Based on user portraits, a product recommendation model is built using machine learning algorithms. The user portrait is used as the model input. Combined with the product's public opinion popularity and store activity data in the region, the product recommendation model outputs a recommended list of products. Collect behavioral data and real-time data of store activities, dynamically adjust the recommendation list, and record user feedback on recommended products.

2. The commodity recommendation method based on big data analysis according to claim 1, characterized in that: The specific steps of collecting the user behavior data on the platform and the store data in the area where the user is located and collecting the public opinion data of the product from the social platform and the comment area are as follows: Combine the IP address and GPS location data of the user's device to initially obtain the user's geographic location; Bind the user ID and geographic location to initially create records; Obtain user behavior data on the platform through the product recommendation platform; According to the user's geographic location, filter the stores in the user's area from the store database; Collect product-related public opinion data from social platforms and comment areas.

3. The commodity recommendation method based on big data analysis according to claim 2, characterized in that: The behavior data, store data and public opinion data are integrated to form a recommendation data set and preprocessed, specifically: Integrate behavioral data, store data, and public opinion data to form a recommended data set; Use a hash table to generate a unique hash value for each piece of data in the recommended data set, compare the hash values ​​to see if they are repeated, and delete the duplicate data after hash collision detection; Use the interquartile range (IQR) to detect outliers in the recommended data set, correct the outliers, set an outlier threshold, and correct the data that exceeds the outlier threshold by the mean of similar data; Through the K nearest neighbor algorithm, the value of the missing data is predicted based on the values ​​of the k neighboring points that are most similar to the current data point, and then the missing data is completed through the KNN interpolation method.

4. The commodity recommendation method based on big data analysis according to claim 3, characterized in that: The specific steps of extracting behavior features from behavior data, extracting store features from store data, and extracting public opinion features from public opinion data are as follows: Aggregate and analyze behavioral data, count the frequency of users' behaviors within a certain time window, and extract the distribution characteristics of behavior types; Count the number of products on shelves and their classification distribution in stores in the user's area, and extract matching features between product categories and inventory; Count the number of times a product is mentioned on social media, extract the popularity features of the product, analyze the review content through the sentiment analysis model, extract the sentiment tendency score of the product, and combine the product popularity and sentiment tendency score to calculate the evaluation features of the product.

5. The commodity recommendation method based on big data analysis according to claim 4, characterized in that: The use of a machine learning algorithm to construct a product recommendation model refers to using a DeepFM model based on deep learning as a product recommendation model, combining user portraits, product features, and dynamic features to output a user's interest prediction score for the product; The product characteristics mainly refer to the basic information of the product and the characteristics of public opinion popularity; The dynamic feature refers to the real-time update of store activity data and behavior data in the area.

6. The commodity recommendation method based on big data analysis according to claim 5, characterized in that: The user portrait is used as the input of the model, combined with the public opinion popularity of the product and the store activity data in the area, and the product recommendation model outputs a recommended list of products. The specific steps are: Map product features and dynamic features into low-dimensional dense vector space to generate feature embedding representation; For the feature cross-FM part of the product recommendation model, a factorization machine is used to capture the second-order interaction relationship between the input features to obtain the output of the FM part ; For the nonlinear feature learning of the product recommendation model, the Deep part is learned by using a fully connected deep neural network to extract high-order nonlinear feature interactions and obtain the output of the Deep part. ; The outputs of the FM part and the Deep part are weighted and fused to obtain the final interest prediction score. ; Predict ratings based on interest , sort all the items' interest prediction scores in descending order, and select the one with the highest score Generate a recommended list of products , and the recommended list Display to the user.

7. The commodity recommendation method based on big data analysis according to claim 6, characterized in that: The dynamically adjusting the recommendation list refers to dynamically calculating the recommendation priority score of each commodity and adjusting the recommendation sequence based on the recommendation priority score.

8. A product recommendation system based on big data analysis, based on the product recommendation method based on big data analysis according to any one of claims 1 to 7, characterized in that: Including collection module, feature extraction module, product recommendation module, adjustment feedback module and display module: The acquisition module is responsible for collecting user-related data, preprocessing and integrating the data, and generating a recommended data set; The feature extraction module is responsible for extracting key features from the collected data and generating user portraits to provide input for the recommendation model; The product recommendation module builds a recommendation model based on user portraits and product features, predicts the user's interest score for the product and generates a recommendation list; The adjustment feedback module is used to collect user behavior data and store activity data in real time, dynamically adjust the recommendation list, and record and analyze user feedback; The display module is used to display the recommendation results to the user in a user-friendly manner and provide interactive functions.

Images