Commodity recommendation method, device and equipment and storage medium

Abstract

The invention discloses a commodity recommendation method and device, equipment and a storage medium, and relates to the technical field of computers, and the method comprises the steps: obtaining e-commerce data based on a preset data receiving interface, and processing the e-commerce data through employing a preset data preprocessing rule, so as to obtain processed data; determining a user portrait and a commodity feature vector based on the processed data, and randomly generating an initial strategy set according to the user portrait and the commodity feature vector; iterating the initial strategy set by using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model comprises a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a leapfrog algorithm, an association rule mining algorithm and a clustering analysis algorithm; and generating a corresponding commodity recommendation list by using the target strategy set so as to display the commodity recommendation list on the target user interface. Therefore, the commodity recommendation accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a commodity recommendation method, device, equipment and storage medium. Background Art

[0002] With the rapid development of e-commerce, the variety of goods is becoming increasingly rich, and the information overload problem faced by users is becoming increasingly prominent. Although traditional product recommendation systems have alleviated this dilemma to a certain extent, they are often difficult to deeply explore and accurately capture users' personalized needs and potential interests due to their relatively simple algorithm models, resulting in a large deviation between the recommendation results and the actual expectations of users, and are not accurate enough.

[0003] Therefore, how to improve the accuracy of product recommendations is a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a product recommendation method, device, equipment and storage medium, which can improve the accuracy of product recommendations. The specific scheme is as follows:

[0005] In a first aspect, the present application provides a product recommendation method, comprising:

[0006] Acquire e-commerce data based on a preset data receiving interface, and process the e-commerce data using a preset data preprocessing rule to obtain processed data;

[0007] Determine a user portrait and a product feature vector based on the processed data, and randomly generate an initial strategy set according to the user portrait and the product feature vector;

[0008] The initial strategy set is iterated by using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm and a cluster analysis algorithm;

[0009] The target policy set is used to generate a corresponding commodity recommendation list, so as to display the commodity recommendation list on a target user interface.

[0010] Optionally, the acquiring e-commerce data based on a preset data receiving interface includes:

[0011] Obtaining explicit feedback data and implicit feedback data of target users from the target e-commerce platform based on a preset application programming interface;

[0012] determining user behavior data based on the explicit feedback data and the implicit feedback data;

[0013] Obtain price information, brand identifiers, inventory turnover rates, and historical promotion sensitivity coefficients of each product on the target e-commerce platform to determine product structured attribute data;

[0014] Extracting semantic embedding vectors of product text description information of each product on the target e-commerce platform using a pre-trained language model based on the Transformer architecture;

[0015] Using a deep convolutional neural network based on a residual network architecture to extract deep convolutional features of the main visual image of each product on the target e-commerce platform;

[0016] Determining unstructured attribute data of a commodity based on the semantic embedding vector and the deep convolutional features;

[0017] Determining commodity attribute data based on the commodity structured attribute data and the commodity unstructured attribute data;

[0018] E-commerce data is determined based on the user behavior data and the commodity attribute data.

[0019] Optionally, determining the user portrait and the product feature vector based on the processed data includes:

[0020] Extracting the target user's gender information, age information, and region information from the processed data to determine the user's basic identity information;

[0021] The TF-IDF algorithm is used to adjust the weight of the user's historical search terms, and the attention mechanism and the adjusted weighted user historical search terms are used to analyze the user's interest change trend to determine the user's interest preference information;

[0022] Determine a corresponding gamma distribution function based on each order amount, and determine a price sensitivity index and a luxury preference coefficient of a target user based on the gamma distribution function to determine user spending power information;

[0023] Analyze the target user's interactive information in each comment area on the target e-commerce platform through a page ranking algorithm, and generate an influence propagation weight value based on the interactive information to determine the user's social influence information;

[0024] Determine a user portrait based on the user's basic identity information, the user's interest preference information, the user's consumption power information, and the user's social influence information;

[0025] Extracting initial numerical features and initial category features of each commodity from the processed data;

[0026] Performing a preset feature transformation on the initial numerical feature to obtain a target numerical feature, and determining a target categorical feature based on the initial categorical feature and a target encoding technology;

[0027] The target numerical features and the target categorical features are subjected to high-order feature interaction through a field-aware decomposition machine model to generate a product feature vector.

[0028] Optionally, the randomly generating an initial strategy set according to the user portrait and the product feature vector further includes:

[0029] Constructing a fitness function suitable for evaluating the product recommendation strategy, and using the fitness function to evaluate each strategy in the initial strategy set to obtain a fitness evaluation result;

[0030] Among them, the fitness function suitable for evaluating the product recommendation strategy is a function constructed based on user behavior information and product distribution dispersion information on the target e-commerce platform.

[0031] Optionally, the using the hybrid bionic algorithm model to iterate the initial strategy set to obtain a target strategy set includes:

[0032] Initializing the parameters of each algorithm in the hybrid bionic algorithm model to complete the parameter initialization operation;

[0033] Executing a first preset operation on the initial strategy set based on the fitness evaluation result by the particle swarm algorithm in the hybrid bionic algorithm model;

[0034] Performing a second preset operation on the initial strategy set after performing the first preset operation by the genetic algorithm in the hybrid bionic algorithm model;

[0035] Performing a third preset operation on the initial strategy set after performing the second preset operation by the ant colony algorithm in the hybrid bionic algorithm model;

[0036] Performing a fourth preset operation on the initial strategy set after performing the third preset operation by the hybrid frog leaping algorithm in the hybrid bionic algorithm model;

[0037] Performing a fifth preset operation on the initial strategy set after performing the fourth preset operation by using the association rule mining algorithm and the cluster analysis algorithm in the hybrid bionic algorithm model;

[0038] The first preset operation, the second preset operation, the third preset operation, the fourth preset operation and the fifth preset operation are repeatedly performed based on a preset number of iterations to obtain a target policy set.

[0039] Optionally, after displaying the product recommendation list on the target user interface, the method further includes:

[0040] Real-time monitoring of current user behavior data and commodity market trends;

[0041] When it is monitored that the change in the user behavior data and / or the commodity market trend is greater than a preset change threshold, the process jumps back to the step of acquiring e-commerce data based on the preset data receiving interface, and processes the e-commerce data using preset data preprocessing rules to obtain processed data.

[0042] In a second aspect, the present application provides a product recommendation device, comprising:

[0043] A data processing module, used to obtain e-commerce data based on a preset data receiving interface, and process the e-commerce data using a preset data preprocessing rule to obtain processed data;

[0044] A strategy set generation module, used to determine a user portrait and a product feature vector based on the processed data, and randomly generate an initial strategy set according to the user portrait and the product feature vector;

[0045] A strategy set iteration module, used for iterating the initial strategy set using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm and a cluster analysis algorithm;

[0046] The commodity display module is used to generate a corresponding commodity recommendation list using the target strategy set, so as to display the commodity recommendation list on the target user interface.

[0047] In a third aspect, the present application provides an electronic device, including:

[0048] Memory, used to store computer programs;

[0049] The processor is used to execute the computer program to implement the aforementioned product recommendation method.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the aforementioned product recommendation method is implemented.

[0051] In the present application, e-commerce data is obtained based on a preset data receiving interface, and the e-commerce data is processed using preset data preprocessing rules to obtain processed data; user portraits and product feature vectors are determined based on the processed data, and an initial strategy set is randomly generated based on the user portraits and the product feature vectors; the initial strategy set is iterated using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm, and a cluster analysis algorithm; the target strategy set is used to generate a corresponding product recommendation list to display the product recommendation list on the target user interface. As can be seen from the above, the present application first obtains e-commerce data through a preset data receiving interface, and then processes these data using preset data preprocessing rules to obtain processed data. Next, the user portrait and product feature vector are determined based on the processed data, and then the initial strategy set is randomly generated in combination with the user portrait and the product feature vector. Afterwards, the initial strategy set is iteratively optimized using a hybrid bionic algorithm model that includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm, and a cluster analysis algorithm, and finally the target strategy set is obtained. Finally, a corresponding product recommendation list is generated according to the target strategy set and displayed on the target user interface. In this way, this application can improve the accuracy of product recommendations on e-commerce platforms, thereby enhancing the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0053] Figure 1 A flowchart of a product recommendation method disclosed in this application;

[0054] Figure 2 A flowchart of a specific product recommendation method disclosed in this application;

[0055] Figure 3 This is a schematic diagram of the structure of a commodity recommendation device disclosed in this application;

[0056] Figure 4 This is a structural diagram of an electronic device disclosed in this application. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] Although traditional product recommendation systems have alleviated this dilemma to a certain extent, they are often difficult to deeply explore and accurately capture users' personalized needs and potential interests because of their relatively simple algorithm models, resulting in a large deviation between the recommendation results and the actual expectations of users, which is not accurate enough. To this end, the present application provides a product recommendation method, device, equipment and storage medium that can improve the accuracy of product recommendations.

[0059] See also Figure 1 As shown, an embodiment of the present invention discloses a commodity recommendation method, comprising:

[0060] Step S11: acquiring e-commerce data based on a preset data receiving interface, and processing the e-commerce data using preset data preprocessing rules to obtain processed data.

[0061] In this embodiment, first, on the one hand, data needs to be obtained from the target e-commerce platform through a preset data receiving interface. Specifically, explicit feedback data and implicit feedback data of the target user are obtained from the target e-commerce platform based on a preset application programming interface. Among them, explicit feedback data includes data such as user ratings and comments on products that directly express their views and preferences on products, while implicit feedback data covers users' browsing behaviors such as browsed product pages, dwell time, etc., click behaviors, and purchase behaviors, etc. These behaviors indirectly reflect users' interests and preferences. Therefore, through the comprehensive analysis and processing of explicit feedback data and implicit feedback data, user behavior data can be determined more accurately.

[0062] In terms of obtaining commodity-related data, it is necessary to obtain multi-dimensional information of each commodity from the target e-commerce platform. Specifically, in this embodiment, the price information, brand identifier, inventory turnover rate, and historical promotion sensitivity coefficient of each commodity on the target e-commerce platform are obtained. Among them, the price information of each commodity can not only include the current selling price of the commodity, but also cover the historical price fluctuations to analyze the impact of price on user purchasing decisions. In addition, since brands often play an important role in users' purchasing choices, it is also necessary to obtain brand identifiers. The inventory turnover rate reflects the sales speed and popularity of the commodity, and commodities with high inventory turnover rates are usually more popular in the market. In addition, the historical promotion sensitivity coefficient can measure the performance of commodities in promotional activities, understand which commodities are more sensitive to promotional activities, and help to formulate more effective recommendation strategies. Based on the information obtained, the structured attribute data of the commodity can be determined.

[0063] For the unstructured attribute data of commodities, firstly, a pre-trained language model based on the Transformer (a sequence model based on the attention mechanism) architecture is used to extract the semantic embedding vectors of the commodity text description information of each commodity on the target e-commerce platform. Among them, the commodity text description information usually contains the detailed characteristics, functions, and usage methods of the commodity, and the Transformer architecture can effectively capture the semantic information in the text and extract the semantic embedding vectors of the commodity text description information. These vectors convert the text information into numerical form, which is convenient for computer processing and analysis.

[0064] At the same time, a deep convolutional neural network based on a residual network architecture is used to extract deep convolutional features of the main visual images of each product on the target e-commerce platform. It should be noted that the main visual image of a product is the first visual information that users see when browsing a product. The residual network architecture can effectively extract deep convolutional features of the image, which contain key visual information of the image, such as color, shape, texture, etc. Through the comprehensive analysis and processing of semantic embedding vectors and deep convolutional features, the unstructured attribute data of the product can be determined.

[0065] Furthermore, after obtaining the user behavior data, the commodity structured attribute data and the commodity unstructured attribute data, the commodity attribute data is determined based on the commodity structured attribute data and the commodity unstructured attribute data. Next, the e-commerce data is determined based on the user behavior data and the commodity attribute data.

[0066] After obtaining the e-commerce data, it is processed using the preset data preprocessing rules. The preset data preprocessing rules include data cleaning, data conversion, data normalization and other operations. Among them, data cleaning mainly removes noise and duplicate data from the data, processes missing values ​​and outliers, etc., to ensure the accuracy and completeness of the data. Data conversion is to convert the data into a format suitable for analysis and model training, such as converting categorical data into numerical data. Data normalization maps the value range of the data to a specific interval to eliminate the dimensional differences between different features and improve the training effect and accuracy of the model. Through the application of these preset data preprocessing rules, the e-commerce data is processed to obtain processed data.

[0067] Step S12: determining a user portrait and a product feature vector based on the processed data, and randomly generating an initial strategy set according to the user portrait and the product feature vector.

[0068] In this embodiment, after the acquisition and preprocessing of the e-commerce data are completed and the processed data is obtained, the gender information, age information and region information of the target user are extracted from the processed data to determine the basic identity information of the user.

[0069] In a specific implementation, the weight of the user's historical search terms is adjusted by the TF-IDF algorithm (Term Frequency-Inverse Document Frequency, a weighted algorithm for information retrieval and text mining), and the attention mechanism and the user's historical search terms after weight adjustment are used to analyze the user's interest change trend to determine the user's interest preference information. Among them, the TF-IDF algorithm can highlight words with high discrimination and importance in the user's search history. The attention mechanism can focus on the key parts of the user's search terms and accurately capture the dynamic changes of the user's interests.

[0070] Next, since the price sensitivity index reflects the user's sensitivity to price changes, and the luxury preference coefficient reflects the user's consumption tendency for luxury goods, this has a certain guiding role in the pricing and category selection of personalized recommended goods. Therefore, in this embodiment, the corresponding gamma distribution function can be determined based on the order amount on the target e-commerce platform, and the price sensitivity index and luxury preference coefficient of the target user can be determined based on the gamma distribution function to determine the user's consumption power information.

[0071] In addition, the interactive information of the target user in each comment area on the target e-commerce platform is analyzed through the page ranking algorithm, and the influence propagation weight value is generated based on the interactive information to determine the user's social influence information. Specifically, the interactive information in the comment area includes the user's views on the product, communication with other users, etc. Based on this interactive information, the influence propagation weight value is generated to determine the user's social influence information.

[0072] After obtaining the user's basic identity information, user interest preference information, user consumption power information, and user social influence information, further determine the user portrait based on the user's basic identity information, user interest preference information, user consumption power information, and user social influence information.

[0073] In a specific implementation, initial numerical features and initial category features of each commodity are extracted from the processed data. It should be noted that the initial numerical features may include attributes such as the price, sales volume, inventory, etc. of the commodity that can be directly represented by numerical values, while the initial category features may include classification attributes such as the brand and category of the commodity.

[0074] Furthermore, the initial numerical features are transformed into preset features to obtain target numerical features, and the target categorical features are determined based on the initial categorical features and the target encoding technology. Furthermore, the target numerical features and the target categorical features are subjected to high-order feature interaction through the field-aware decomposition machine model to generate a product feature vector. After determining the user profile and product feature vector, an initial strategy set is randomly generated based on this information. The initial strategy set contains a series of initial product recommendation strategies. Obviously, these recommendation strategies are randomly generated based on the feature information of users and products.

[0075] In addition, in this embodiment, after the strategy set is generated, a fitness function suitable for evaluating the product recommendation strategy is constructed, and each strategy in the initial strategy set is evaluated using the fitness function to obtain a fitness evaluation result. In addition, the obtained fitness evaluation result will provide a certain basis for the subsequent iteration of the initial strategy set. Among them, the fitness function suitable for evaluating the product recommendation strategy is a function constructed based on user behavior information and product distribution dispersion information on the target e-commerce platform.

[0076] Step S13, using a hybrid bionic algorithm model to iterate the initial strategy set to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm and a cluster analysis algorithm.

[0077] Before starting the iteration, the parameters of each algorithm in the hybrid bionic algorithm model need to be initialized to complete the parameter initialization operation. For the particle swarm algorithm, it is necessary to set the appropriate number of particles, speed range and learning factor. For the genetic algorithm, it is necessary to initialize the population size, crossover probability and mutation probability. The ant colony algorithm needs to set the pheromone volatility coefficient and the number of ants. In addition, the association rule mining algorithm and the cluster analysis algorithm also need to set corresponding parameters, such as the minimum support and minimum confidence of association rule mining, the number of clusters of cluster analysis, etc.

[0078] In the process of iterating the initial strategy set, first, the particle swarm algorithm in the hybrid bionic algorithm model performs the first preset operation on the initial strategy set based on the fitness evaluation result. It should be noted that each particle in the particle swarm algorithm represents a recommendation strategy, and its position in the search space corresponds to the specific parameter combination of the recommendation strategy. In each iteration, the particle updates its speed and position according to its own historical optimal position (that is, the parameter combination corresponding to the optimal fitness of the recommendation strategy represented by the particle in the previous iteration) and the global optimal position (the parameter combination of the recommendation strategy with the highest fitness among all particles at present). Through the global search capability of the particle swarm algorithm, the particles are continuously guided to approach a better recommendation strategy, so that the recommendation strategy in the initial strategy set is gradually optimized to provide users with product recommendations that are more in line with their personal interests.

[0079] Then, the genetic algorithm in the hybrid bionic algorithm model performs the second preset operation on the initial strategy set after the first preset operation. Among them, the genetic algorithm simulates the biological evolution process and evolves the recommended strategy through selection, crossover and mutation operations. The selection operation can select recommended strategies with higher fitness from the current strategy set as parents. These parents will have more opportunities to pass the parameters of their own recommended strategies to the next generation. The crossover operation exchanges some parameters of the parent recommended strategy to generate new offspring recommended strategies. The mutation operation randomly changes the parameters of the recommended strategy with a certain probability, introduces new diversity to the strategy set, and avoids the algorithm from falling into the local optimum. Through the evolution process of the genetic algorithm, the overall quality of the strategy set is further improved.

[0080] Then, the ant colony algorithm in the hybrid bionic algorithm model performs a third preset operation on the initial strategy set after the second preset operation. It should be noted that the ant colony algorithm simulates the foraging behavior of ants, and ants leave pheromones on the paths they pass through during the search process. In the scenario of product recommendation, each path represents a recommendation strategy, and the pheromone concentration indicates the pros and cons of the recommendation strategy. Ants select paths based on the pheromone concentration, and the higher the pheromone concentration, the greater the probability of being selected. At the same time, pheromones evaporate over time, which prompts ants to continuously explore new paths. Through the local search capability of the ant colony algorithm, fine adjustments are made on the basis of the existing recommendation strategies to find a better recommendation path. For example, it can further optimize the combination of recommended products within a specific product category or price range.

[0081] Next, the hybrid frog leaping algorithm in the hybrid bionic algorithm model performs the fourth preset operation on the initial strategy set after the third preset operation. The hybrid frog leaping algorithm divides the population into multiple subgroups for local search, and realizes global search through information exchange between subgroups. The frogs in each subgroup represent the recommended strategy, and the frogs in each subgroup search in the local area and update their positions to find a better recommended strategy. The subgroups regularly exchange information and share their optimal solutions. This combination of local search and global search allows the recommended strategy to be finely optimized locally while also being coordinated and improved globally.

[0082] Finally, the fifth preset operation is performed on the initial strategy set after the fourth preset operation is performed through the association rule mining algorithm and the cluster analysis algorithm in the hybrid bionic algorithm model. Among them, the association rule mining algorithm discovers the frequent co-occurrence pattern between commodities by analyzing a large amount of transaction data. This association rule can provide more basis for the recommendation strategy. The cluster analysis algorithm classifies commodities and users according to similarity, classifies users with similar characteristics or purchasing behaviors into one category, and classifies commodities with similar attributes or uses into one category. In the recommendation process, commodities preferred by users in the category can be recommended to the user according to the category to which the user belongs. Through the combination of these two algorithms, the potential relationship between users and commodities can be further mined, and the recommendation strategy can be refined and optimized, so that the recommendation strategy is more in line with the actual needs of users and market laws, thereby generating a more targeted and accurate commodity recommendation list.

[0083] The first preset operation, the second preset operation, the third preset operation, the fourth preset operation and the fifth preset operation are repeatedly performed based on the preset number of iterations. As the iteration proceeds, the recommended strategy set is continuously optimized and the fitness is gradually improved. When the preset number of iterations is reached, the iteration process is terminated, and the strategy set obtained at this time is the target strategy set.

[0084] Step S14: Generate a corresponding commodity recommendation list using the target policy set, so as to display the commodity recommendation list on the target user interface.

[0085] After obtaining the target strategy set, a corresponding product recommendation list is generated for each user according to the recommendation strategy in the target strategy set. Each recommendation strategy contains a series of parameters and rules, such as the category, price range, and sorting rules of the recommended products. Based on these parameters and rules, qualified products are screened from the product database, and a recommendation list is generated according to a certain sorting method.

[0086] In order to ensure the timeliness and accuracy of the recommendation results, the system needs to monitor the current user behavior data and commodity market trends in real time. User behavior data includes users' browsing history, click behavior, purchase history, etc. These data reflect the real-time changes in users' interests and needs. Commodity market trends include commodity price fluctuations, new product launches, promotional activities and other information, which will affect users' purchasing decisions.

[0087] When the change in user behavior data and / or commodity market trends is greater than the preset change threshold, it indicates that the user's interests or market environment have changed significantly. At this time, the process will jump back to the step of obtaining e-commerce data based on the preset data receiving interface, and using the preset data preprocessing rules to process the e-commerce data to obtain the processed data. By re-acquiring and processing the data, updating the user portrait and commodity feature vector, the hybrid bionic algorithm model is used again to optimize the recommendation strategy and generate a new commodity recommendation list. This ensures that changes can be adapted in a timely manner, providing users with commodity recommendations that always meet their needs.

[0088] Furthermore, in this embodiment, the generated product recommendation list is displayed on the target user interface. The design of the user interface should follow the principles of intuitiveness and friendliness, and use lists, carousels, grids and other forms to display recommended products to meet the browsing habits of different users. At the same time, a personalized recommendation reason or label is provided for each recommended product to enhance the user's trust and satisfaction with the recommendation results.

[0089] As can be seen from the above, the present application first obtains e-commerce data through a preset data receiving interface, and then processes these data using preset data preprocessing rules to obtain processed data. Next, the user portrait and product feature vector are determined based on the processed data, and then the initial strategy set is randomly generated in combination with the user portrait and the product feature vector. Afterwards, the initial strategy set is iteratively optimized using a hybrid bionic algorithm model that includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm, and a cluster analysis algorithm, and finally the target strategy set is obtained. Finally, a corresponding product recommendation list is generated according to the target strategy set and displayed on the target user interface. In this way, the present application can improve the accuracy of product recommendations on the e-commerce platform, thereby enhancing the user experience.

[0090] Combine the following Figure 2 The schematic diagram shown specifically illustrates the technical solution of the embodiment of the present application.

[0091] Specifically, first, we need to conduct a demand analysis. Taking the e-commerce platform as an example, we need to make it clear that the system needs to implement personalized recommendations, and accurately push products based on the user's past browsing and purchasing records and preferences on the platform. For example, for users who frequently purchase sports equipment, we can recommend new sports shoes or sportswear to them. At the same time, we need to ensure diverse recommendations to prevent the recommended products from being limited to a certain category. Real-time performance is also crucial. Once a user has new browsing or purchasing behavior on the platform, the system can quickly capture and adjust the recommendation results.

[0092] Next is data collection, processing and storage. Obtain data through the data interface of the e-commerce platform, or use crawler technology under the premise of legality and compliance to collect data generated by users' browsing, clicking, purchasing, and evaluation on the platform, as well as information such as the price, brand, category, sales volume, and evaluation of the product. Clean the collected data, remove duplicate data, process missing values, and detect and correct outliers to ensure that the data is accurate and complete. Then extract key features from the cleaned data, build user portraits containing information such as age, gender, region, and interest preferences, and product feature vectors covering price, brand, evaluation level, keywords, etc., and finally store the data in a suitable database for subsequent analysis and processing.

[0093] Then, a hybrid bionic algorithm model is constructed. Initialize the algorithm parameters first. The particle swarm algorithm needs to set the number of particles, speed range, and learning factor; the genetic algorithm needs to set the population size, crossover probability, and mutation probability; the ant colony algorithm needs to set the pheromone volatility coefficient and the number of ants, as well as the relevant parameters of association rule mining and cluster analysis. Then, the initial population (that is, the initial strategy set) is randomly generated based on the user portrait and the product feature vector. Each strategy contains a list of recommended products. Design the fitness function to evaluate the advantages and disadvantages of the recommendation strategy from the dimensions of user satisfaction (such as click-through rate, conversion rate, etc.), product diversity, and recommendation novelty. Then, the algorithm enters the iterative optimization stage. The particle swarm algorithm updates the particle speed and position according to the fitness evaluation results, and approaches the global optimal solution. The genetic algorithm evolves the population through selection, crossover and mutation operations to generate a new recommendation strategy set. The ant colony algorithm uses the pheromone mechanism to simulate ant foraging and search for better recommendation paths locally. The hybrid frog leaping algorithm divides the population into multiple subgroups for local search, and realizes global search through information exchange between subgroups, and further optimizes on the basis of the genetic algorithm or particle swarm algorithm. The association rule mining analyzes the transaction data to find the frequent co-occurrence pattern of commodities, and the cluster analysis clusters commodities and users according to the similarity metric to provide a basis for recommendation. These steps are repeated until the preset number of iterations is reached or the stop condition is met.

[0094] Finally, based on the optimal recommendation strategy (i.e., the target strategy set) output by the hybrid bionic algorithm model, a personalized product recommendation list is generated for the user. User behavior and market trends are monitored in real time, user feedback data such as clicks, purchases, and reviews are collected, and the recommendation strategy is adjusted dynamically accordingly. If there are major changes in user interests or market trends, the global optimization of the algorithm model is re-triggered. An intuitive and friendly user interface is designed to display the recommendation list in the form of a list, a carousel, a grid, etc., and personalized reasons or labels for the recommendation are provided. At the same time, user feedback and behavior data are collected through the user interface, and the performance of the recommendation system and user satisfaction are analyzed and evaluated to provide a basis for the continuous adjustment and optimization of the recommendation strategy.

[0095] Accordingly, see Figure 3 As shown, the embodiment of the present application provides a commodity recommendation device, including:

[0096] The data processing module 11 is used to obtain e-commerce data based on a preset data receiving interface, and process the e-commerce data using a preset data preprocessing rule to obtain processed data;

[0097] A strategy set generation module 12, configured to determine a user portrait and a product feature vector based on the processed data, and randomly generate an initial strategy set according to the user portrait and the product feature vector;

[0098] A strategy set iteration module 13 is used to iterate the initial strategy set using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm, and a cluster analysis algorithm;

[0099] The commodity display module 14 is used to generate a corresponding commodity recommendation list using the target policy set, so as to display the commodity recommendation list on the target user interface.

[0100] As can be seen from the above, the present application first obtains e-commerce data through a preset data receiving interface, and then processes these data using preset data preprocessing rules to obtain processed data. Next, the user portrait and product feature vector are determined based on the processed data, and then the initial strategy set is randomly generated in combination with the user portrait and the product feature vector. Afterwards, the initial strategy set is iteratively optimized using a hybrid bionic algorithm model that includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm, and a cluster analysis algorithm, and finally the target strategy set is obtained. Finally, a corresponding product recommendation list is generated according to the target strategy set and displayed on the target user interface. In this way, the present application can improve the accuracy of product recommendations on the e-commerce platform, thereby enhancing the user experience.

[0101] In some specific implementations, the data processing module 11 specifically includes:

[0102] A data acquisition unit, used for acquiring explicit feedback data and implicit feedback data of a target user from a target e-commerce platform based on a preset application programming interface;

[0103] A first data determining unit, configured to determine user behavior data based on the explicit feedback data and the implicit feedback data;

[0104] An information acquisition unit, used to acquire price information, brand identifier, inventory turnover rate, and historical promotion sensitivity coefficient of each product on the target e-commerce platform to determine product structured attribute data;

[0105] A vector extraction unit, used to extract semantic embedding vectors of product text description information of each product on the target e-commerce platform using a pre-trained language model based on a Transformer architecture;

[0106] A feature extraction unit, used to extract deep convolutional features of the main visual image of each product on the target e-commerce platform using a deep convolutional neural network based on a residual network architecture;

[0107] A second data determination unit, configured to determine commodity unstructured attribute data based on the semantic embedding vector and the deep convolutional feature;

[0108] a third data determination unit, configured to determine commodity attribute data based on the commodity structured attribute data and the commodity unstructured attribute data;

[0109] The fourth data determination unit is used to determine the e-commerce data based on the user behavior data and the commodity attribute data.

[0110] In some specific implementations, the policy set generation module 12 specifically includes:

[0111] An identity information determination unit, used to extract the gender information, age information and region information of the target user from the processed data to determine the basic identity information of the user;

[0112] A preference information determination unit, used to adjust the weight of the user's historical search terms through the TF-IDF algorithm, and analyze the user's interest change trend by using the attention mechanism and the user's historical search terms after the weight adjustment, so as to determine the user's interest preference information;

[0113] A consumption power information determination unit, used to determine a corresponding gamma distribution function based on each order amount, and determine a price sensitivity index and a luxury preference coefficient of a target user based on the gamma distribution function to determine the user's consumption power information;

[0114] A social influence determination unit is used to analyze the interactive information of the target user in each comment area on the target e-commerce platform through a page ranking algorithm, and generate an influence propagation weight value based on the interactive information to determine the user's social influence information;

[0115] A portrait determination unit, configured to determine a user portrait based on the user basic identity information, the user interest preference information, the user consumption power information, and the user social influence information;

[0116] An initial feature determination unit, used to extract initial numerical features and initial category features of each commodity from the processed data;

[0117] a target feature determination unit, configured to perform a preset feature transformation on the initial numerical feature to obtain a target numerical feature, and determine a target category feature based on the initial category feature and a target encoding technology;

[0118] The vector generation unit is used to perform high-order feature interaction on the target numerical feature and the target categorical feature through a field-aware decomposition machine model to generate a product feature vector.

[0119] In some specific implementations, the policy set generation module 12 specifically further includes:

[0120] A strategy evaluation unit, used to construct a fitness function suitable for evaluating the product recommendation strategy, and use the fitness function to evaluate each strategy in the initial strategy set to obtain a fitness evaluation result;

[0121] Among them, the fitness function suitable for evaluating the product recommendation strategy is a function constructed based on user behavior information and product distribution dispersion information on the target e-commerce platform.

[0122] In some specific implementations, the policy set iteration module 13 specifically includes:

[0123] A parameter initialization unit, used to initialize the parameters of each algorithm in the hybrid bionic algorithm model to complete the parameter initialization operation;

[0124] A first operation execution unit, configured to execute a first preset operation on the initial strategy set based on the fitness evaluation result by using the particle swarm algorithm in the hybrid bionic algorithm model;

[0125] A second operation execution unit, configured to execute a second preset operation on the initial strategy set after executing the first preset operation through the genetic algorithm in the hybrid bionic algorithm model;

[0126] A third operation execution unit, configured to execute a third preset operation on the initial strategy set after executing the second preset operation through the ant colony algorithm in the hybrid bionic algorithm model;

[0127] a fourth operation execution unit, configured to execute a fourth preset operation on the initial strategy set after executing the third preset operation by using the hybrid frog leaping algorithm in the hybrid bionic algorithm model;

[0128] a fifth operation execution unit, configured to execute a fifth preset operation on the initial strategy set after executing the fourth preset operation by using the association rule mining algorithm and the cluster analysis algorithm in the hybrid bionic algorithm model;

[0129] The target strategy determination unit is used to repeatedly perform the first preset operation, the second preset operation, the third preset operation, the fourth preset operation and the fifth preset operation based on a preset number of iterations to obtain a target strategy set.

[0130] In some specific implementations, the product display module 14 further includes:

[0131] Data monitoring unit, used to monitor current user behavior data and commodity market trends in real time;

[0132] The data processing unit is used to jump back to the step of acquiring e-commerce data based on the preset data receiving interface and processing the e-commerce data using preset data preprocessing rules to obtain processed data when it is monitored that the change in the user behavior data and / or the commodity market trend is greater than a preset change threshold.

[0133] Furthermore, the present application also discloses an electronic device. Figure 4 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be regarded as any limitation on the scope of use of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input and output interface 25 and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the commodity recommendation method disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment can specifically be an electronic computer.

[0134] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0135] In addition, the memory 22 as a carrier for resource storage may be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon may include an operating system 221, a computer program 222, etc., and the storage method may be temporary storage or permanent storage.

[0136] The operating system 221 is used to manage and control the hardware devices and computer program 222 on the electronic device 20, and can be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program that can be used to complete the commodity recommendation method performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 can further include a computer program that can be used to complete other specific tasks.

[0137] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein the computer program, when executed by a processor, implements the aforementioned disclosed commodity recommendation method. For the specific steps of the method, reference may be made to the corresponding contents disclosed in the aforementioned embodiments, and no further description will be given here.

[0138] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0139] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0140] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0141] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0142] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A product recommendation method, characterized in that: include: Acquire e-commerce data based on a preset data receiving interface, and process the e-commerce data using a preset data preprocessing rule to obtain processed data; Determine a user portrait and a product feature vector based on the processed data, and randomly generate an initial strategy set according to the user portrait and the product feature vector; The initial strategy set is iterated by using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm and a cluster analysis algorithm; The target policy set is used to generate a corresponding commodity recommendation list, so as to display the commodity recommendation list on a target user interface.

2. The commodity recommendation method according to claim 1, characterized in that: The obtaining of e-commerce data based on a preset data receiving interface includes: Obtaining explicit feedback data and implicit feedback data of target users from the target e-commerce platform based on a preset application programming interface; determining user behavior data based on the explicit feedback data and the implicit feedback data; Obtain price information, brand identifiers, inventory turnover rates, and historical promotion sensitivity coefficients of each product on the target e-commerce platform to determine product structured attribute data; Extracting semantic embedding vectors of product text description information of each product on the target e-commerce platform using a pre-trained language model based on the Transformer architecture; Using a deep convolutional neural network based on a residual network architecture to extract deep convolutional features of the main visual image of each product on the target e-commerce platform; Determining unstructured attribute data of a commodity based on the semantic embedding vector and the deep convolutional features; Determining commodity attribute data based on the commodity structured attribute data and the commodity unstructured attribute data; E-commerce data is determined based on the user behavior data and the commodity attribute data.

3. The commodity recommendation method according to claim 1, characterized in that: The determining of the user portrait and the commodity feature vector based on the processed data includes: Extracting the target user's gender information, age information, and region information from the processed data to determine the user's basic identity information; The TF-IDF algorithm is used to adjust the weight of the user's historical search terms, and the attention mechanism and the adjusted weighted user historical search terms are used to analyze the user's interest change trend to determine the user's interest preference information; Determine a corresponding gamma distribution function based on each order amount, and determine a price sensitivity index and a luxury preference coefficient of a target user based on the gamma distribution function to determine user spending power information; Analyze the target user's interactive information in each comment area on the target e-commerce platform through a page ranking algorithm, and generate an influence propagation weight value based on the interactive information to determine the user's social influence information; Determine a user portrait based on the user's basic identity information, the user's interest preference information, the user's consumption power information, and the user's social influence information; Extracting initial numerical features and initial category features of each commodity from the processed data; Performing a preset feature transformation on the initial numerical feature to obtain a target numerical feature, and determining a target categorical feature based on the initial categorical feature and a target encoding technology; The target numerical features and the target categorical features are subjected to high-order feature interaction through a field-aware decomposition machine model to generate a product feature vector.

4. The commodity recommendation method according to claim 2, characterized in that: The randomly generating an initial strategy set according to the user portrait and the commodity feature vector also includes: Constructing a fitness function suitable for evaluating the product recommendation strategy, and using the fitness function to evaluate each strategy in the initial strategy set to obtain a fitness evaluation result; Among them, the fitness function suitable for evaluating the product recommendation strategy is a function constructed based on user behavior information and product distribution dispersion information on the target e-commerce platform.

5. The commodity recommendation method according to claim 4, characterized in that: The hybrid bionic algorithm model is used to iterate the initial strategy set to obtain a target strategy set, including: Initializing the parameters of each algorithm in the hybrid bionic algorithm model to complete the parameter initialization operation; Executing a first preset operation on the initial strategy set based on the fitness evaluation result by the particle swarm algorithm in the hybrid bionic algorithm model; Performing a second preset operation on the initial strategy set after performing the first preset operation by the genetic algorithm in the hybrid bionic algorithm model; Performing a third preset operation on the initial strategy set after performing the second preset operation by the ant colony algorithm in the hybrid bionic algorithm model; Performing a fourth preset operation on the initial strategy set after performing the third preset operation by the hybrid frog leaping algorithm in the hybrid bionic algorithm model; Performing a fifth preset operation on the initial strategy set after performing the fourth preset operation through the association rule mining algorithm and the cluster analysis algorithm in the hybrid bionic algorithm model; The first preset operation, the second preset operation, the third preset operation, the fourth preset operation and the fifth preset operation are repeatedly performed based on a preset number of iterations to obtain a target policy set.

6. The commodity recommendation method according to any one of claims 1 to 5, characterized in that: After displaying the product recommendation list on the target user interface, the method further includes: Real-time monitoring of current user behavior data and commodity market trends; When it is monitored that the change in the user behavior data and / or the commodity market trend is greater than a preset change threshold, the process jumps back to the step of acquiring e-commerce data based on the preset data receiving interface, and processes the e-commerce data using preset data preprocessing rules to obtain processed data.

7. A commodity recommendation device, characterized in that: include: A data processing module, used to obtain e-commerce data based on a preset data receiving interface, and process the e-commerce data using a preset data preprocessing rule to obtain processed data; A strategy set generation module, used to determine a user portrait and a product feature vector based on the processed data, and randomly generate an initial strategy set according to the user portrait and the product feature vector; A strategy set iteration module, used for iterating the initial strategy set using a hybrid bionic algorithm model to obtain a target strategy set; wherein the hybrid bionic algorithm model includes a particle swarm algorithm, a genetic algorithm, an ant colony algorithm, a frog leaping algorithm, an association rule mining algorithm and a cluster analysis algorithm; The commodity display module is used to generate a corresponding commodity recommendation list using the target strategy set, so as to display the commodity recommendation list on the target user interface.

8. The commodity recommendation device according to claim 7, characterized in that: The strategy set iteration module includes: A parameter initialization unit, used to initialize the parameters of each algorithm in the hybrid bionic algorithm model to complete the parameter initialization operation; A first operation execution unit, configured to execute a first preset operation on the initial strategy set based on the fitness evaluation result by using the particle swarm algorithm in the hybrid bionic algorithm model; A second operation execution unit, configured to execute a second preset operation on the initial strategy set after executing the first preset operation through the genetic algorithm in the hybrid bionic algorithm model; A third operation execution unit, configured to execute a third preset operation on the initial strategy set after executing the second preset operation through the ant colony algorithm in the hybrid bionic algorithm model; a fourth operation execution unit, configured to execute a fourth preset operation on the initial strategy set after executing the third preset operation by using the hybrid frog leaping algorithm in the hybrid bionic algorithm model; a fifth operation execution unit, configured to execute a fifth preset operation on the initial strategy set after executing the fourth preset operation by using the association rule mining algorithm and the cluster analysis algorithm in the hybrid bionic algorithm model; The target strategy determination unit is used to repeatedly perform the first preset operation, the second preset operation, the third preset operation, the fourth preset operation and the fifth preset operation based on a preset number of iterations to obtain a target strategy set.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the commodity recommendation method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: Used to store a computer program; wherein, when the computer program is executed by a processor, the commodity recommendation method according to any one of claims 1 to 6 is implemented.

Images