E-commerce intelligent recommendation system based on user behavior big data

Abstract

The invention relates to the technical field of e-commerce intelligent recommendation, in particular to an e-commerce intelligent recommendation system based on user behavior big data, which comprises a data acquisition and storage unit, a data preprocessing unit, a model construction training unit and a recommendation generation and output unit. The method comprises the steps that data such as user behaviors, equipment, geographic positions and access time are collected and stored in a distributed database, a data preprocessing unit extracts the data, constructs a user behavior sequence and encodes commodity and behavior characteristics, and a model construction training unit obtains user interest vectors by using a long and short-term memory network in combination with an attention mechanism; model parameters are updated with the help of a deep Q network, commodities are screened, sorted and displayed by a recommendation generation and output unit according to interest vectors, meanwhile, a feedback optimization system is collected, recommendation accuracy is improved, user shopping experience is enhanced, and an e-commerce platform is assisted to improve operation efficiency and commodity sales conversion rate.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce intelligent recommendation, and in particular to an e-commerce intelligent recommendation system based on user behavior big data. Background Art

[0002] E-commerce intelligent recommendation is an important technology. With the vigorous development of e-commerce, the number of products on e-commerce platforms has exploded, and it has become increasingly difficult for users to select their desired items from a large number of products. At the same time, e-commerce companies are also facing the challenge of how to improve user retention and promote product sales.

[0003] In today's e-commerce shopping scenarios, users often face the problem of information overload when purchasing products. Traditional recommendation systems fail to distinguish the importance of different behaviors due to overly general data collection methods, resulting in insufficient processing of user behavior data and insufficient analysis and mining capabilities. It is difficult to accurately grasp users' specific preferences for sports shoe brands, styles, functions and prices, and it is impossible to recommend suitable products based on market popularity. It is often the case that the recommended products are neither what users want nor what is currently popular, making it difficult for users to screen through a large number of products and reducing consumers' willingness to buy. In order to solve this technical problem, we provide an e-commerce intelligent recommendation system based on user behavior big data. Summary of the invention

[0004] The purpose of the present invention is to provide an e-commerce intelligent recommendation system based on user behavior big data to solve the problems raised in the above background technology.

[0005] To achieve the above purpose, an e-commerce intelligent recommendation system based on user behavior big data is provided, including a data acquisition and storage unit, a data preprocessing unit, a model building and training unit, and a recommendation generation and output unit;

[0006] The data collection and storage unit deploys embedding points on each page of the e-commerce platform, and the embedding points are used to collect user behavior data and user device information, geographic location and access time, and store the collected behavior data in a distributed database;

[0007] The data preprocessing unit extracts and processes the collected behavior data from the distributed database, and then arranges the user's behavior data in chronological order to construct a user behavior sequence, wherein each sequence contains multiple behavior nodes, each node contains product features and behavior features, and finally encodes the product features and behavior features and converts them into numerical vectors;

[0008] The model construction training unit introduces a long short-term memory network as a core model, takes the encoding vector of the user behavior sequence obtained after preprocessing as input, extracts and learns features through multi-layer long short-term memory network units, outputs the hidden state of each time step, applies an attention mechanism to the hidden state, calculates the attention weight of each time step, obtains the final user interest representation vector by weighted summation, finally defines the state, action and reward, adopts a deep Q network as a reinforcement learning algorithm, selects the optimal recommended action according to the current state, and updates the model parameters according to the reward signal;

[0009] The recommendation generation output unit selects products related to the user's interests from the product library according to the user's interest representation vector, forms a candidate product set, outputs a recommendation strategy, sorts the candidate product set, determines the order of recommended products, and finally displays the sorted recommended product list to the user, while recording the user's feedback on the recommendation results.

[0010] As a further improvement of the technical solution, in the data acquisition and storage unit, an adaptive sampling strategy is adopted when collecting user behavior data by embedding points. The specific steps are as follows:

[0011] Define a set of behavior data importance indicators, where the behavior data importance indicators are the correlation between behavior and purchase conversion, the rarity of behavior, and the timeliness of behavior, and assign an initial weight to each indicator;

[0012] For each user behavior event, its importance score is calculated according to the behavior data importance index and its initial weight, and a dynamic sampling rate is set according to the importance score. Adaptive sampling is performed through the dynamic sampling rate to obtain user behavior data.

[0013] As a further improvement of the technical solution, the specific method for the data preprocessing unit to handle the problem of interruption of the behavior sequence caused by long-term inactivity when constructing the user behavior sequence is as follows:

[0014] A time interval threshold is set. When the time interval between two adjacent behavior nodes in the user behavior sequence is found to be greater than the time interval threshold, the sequence is determined to be interrupted.

[0015] For interrupted sequences, a sequence filling strategy is adopted, that is, analyzing the behavioral characteristics before the interruption. If the previous behavior is concentrated on browsing products in a certain category, several similar products are selected from the user's historical browsing history and inserted into the interruption according to the browsing probability to simulate the user's subsequent behavior.

[0016] As a further improvement of the technical solution, the data preprocessing unit adopts a knowledge graph-based encoding method when encoding product features and behavior features. The specific steps are as follows:

[0017] Construct a knowledge graph in the e-commerce field. The nodes include products, brands, categories, and attributes. The edges represent the relationships between them. For product features, find the corresponding category, brand, and key attribute nodes based on the knowledge graph, and concatenate the embedding vectors of these nodes as the encoding vector of the product features.

[0018] For behavioral features, one-hot encoding is used to map them into corresponding behavior vectors, which are then fused with the product feature encoding vector of the behavior node to obtain the final encoding vector containing product and behavior information.

[0019] As a further improvement of the technical solution, the model building training unit includes an extraction learning module, in which the multi-layer long short-term memory network unit in the extraction learning module introduces a gating mechanism when performing feature extraction and learning, and the specific operations are as follows:

[0020] Based on the input gate, forget gate and output gate of the long short-term memory network, an adaptive adjustment gate is added. For each time step in the input sequence, the input of each gate is obtained through linear transformation;

[0021] The input of the adaptive adjustment gate is to dynamically adjust the focus of feature extraction according to the current input and the previous state.

[0022] As a further improvement of the technical solution, the model construction training unit includes an attention mechanism module. When the attention mechanism module applies the attention mechanism on the hidden state, a multi-head attention mechanism is adopted. The specific calculation method is as follows:

[0023] Set the hidden state sequence, extract the hidden state vector from the hidden state sequence, and map the hidden state vector to multiple subspaces through linear transformation to generate The query vector, key vector, and value vector of each head;

[0024] For each head, calculate the attention score, and get the attention weight through the softmax function, then calculate the context vector of each head, and finally The context vectors of the heads are concatenated and linearly transformed to obtain the final attention output.

[0025] As a further improvement of the technical solution, the model building training unit includes a reinforcement learning module. When the reinforcement learning module uses a deep Q network as a reinforcement learning algorithm, the specific method of defining states, actions and rewards is as follows:

[0026] The state consists of the current user's interest representation vector, the user's current e-commerce page information, and the user's historical behavior statistical characteristics;

[0027] Select a certain number of products from the product library as recommended candidate actions, and each action corresponds to a set of candidate products;

[0028] If the user purchases the recommended product within a certain period of time after seeing it, a positive reward will be given. If the user clicks on the recommended product to enter the details page, a half positive reward will be given. If the user does not respond to the recommended product, a negative reward will be given.

[0029] As a further improvement of the technical solution, the recommendation generation output unit adopts a sorting algorithm that integrates user preferences and product popularity when sorting the candidate product set. The specific steps are as follows:

[0030] For each candidate product, calculate its user preference score, extract the user's preference for the product category and brand from the user behavior data, and obtain the user preference score through weighted summation;

[0031] Calculate the popularity score of the product by counting the number of views and purchases of the product within a certain period of time, summarizing them according to the preset weights, and finally calculating the ranking score of the candidate products based on the user preference score and the product popularity score;

[0032] Collect user feedback on recommendation results, organize the feedback data into structured data form, use association rule mining algorithm to analyze the feedback data, and optimize each unit of the recommendation system based on the analysis results.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] In the e-commerce intelligent recommendation system based on user behavior big data, by constructing user behavior sequences, the interruption problem caused by long-term inactivity can be effectively handled. By analyzing the characteristics of previous behaviors to fill in reasonable sequences, the consistency and accuracy of behavior analysis are guaranteed. The encoding method based on knowledge graph deeply explores the potential connection between products and behavior characteristics, so that feature encoding can more accurately reflect user preferences. The model construction training unit introduces long-term and short-term memory networks, gating mechanisms and multi-head attention mechanisms, which can more effectively capture the dynamic changes of user interests. The reinforcement learning module continuously optimizes the recommendation strategy by reasonably defining states, actions and rewards. The recommendation generation output unit integrates user preferences and product popularity to sort products, and optimizes the system according to user feedback, which improves the accuracy and effectiveness of recommendations, improves user shopping experience, and improves the product sales conversion rate and operational efficiency of e-commerce platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is an overall block diagram of the present invention.

[0036] The meaning of each number in the figure is:

[0037] 1. Data acquisition and storage unit; 2. Data preprocessing unit; 3. Model building and training unit; 31. Extraction learning module; 32. Attention mechanism module; 33. Reinforcement learning module; 4. Recommendation generation and output unit. DETAILED DESCRIPTION

[0038] 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.

[0039] The present invention provides an e-commerce intelligent recommendation system based on user behavior big data, please refer to Figure 1 As shown, it includes a data acquisition and storage unit 1, a data preprocessing unit 2, a model building and training unit 3, and a recommendation generation and output unit 4;

[0040] The data collection and storage unit 1 deploys embedding points on each page of the e-commerce platform, which are used to collect user behavior data as well as user device information, geographic location and access time, and store the collected behavior data in a distributed database.

[0041] In the data collection and storage unit 1, an adaptive sampling strategy is used when collecting user behavior data. The specific steps are as follows:

[0042] Define a set of behavioral data importance indicators ,in The correlation between behavior and purchase conversion is measured by counting the proportion of users who purchased goods within a certain period of time after a certain behavior in historical data. is the rarity of the behavior, that is, the frequency of the behavior in all user behaviors, The timeliness of the behavior. The closer the behavior is to the current time, the higher its importance. Assign an initial weight to each indicator. ,and .

[0043] For each user behavior event , calculate its importance score ; Set dynamic sampling rate based on importance score and set high importance threshold and the importance threshold ,when When the sampling rate is ,when When the sampling rate is ,when When the sampling rate is ,Through this adaptive sampling, we can ensure the collection of key ,behavior data and reduce unnecessary data storage.

[0044] The data preprocessing unit 2 extracts and processes the collected behavior data from the distributed database, and then arranges the user's behavior data in chronological order to construct a user behavior sequence, where each sequence contains multiple behavior nodes, each node contains product features and behavior features, and finally encodes the product features and behavior features and converts them into numerical vectors.

[0045] When constructing the user behavior sequence, the data preprocessing unit 2 handles the problem of behavior sequence interruption caused by long-term inactivity as follows:

[0046] In the user behavior sequence, a long period of inactivity may cause a logical gap in the sequence, which will affect the subsequent model's learning and understanding of the user behavior pattern. Set a time interval threshold , can clearly define the situation of sequence interruption and provide clear judgment criteria for subsequent processing.

[0047] For user behavior sequence ,in Indicates Behavior nodes, each behavior node contains the time when the behavior occurs , when two adjacent behavior nodes are found and satisfy When the sequence is interrupted, it is determined to be a sequence interruption. By setting a clear threshold, the determination of sequence interruption is objective and operational, avoiding the uncertainty of subjective judgment.

[0048] For the sequence determined to be interrupted, analyze the behavior nodes, count the commodity categories involved in these behavior nodes, and analyze them based on the user's historical behavior characteristics. This can make full use of the user's historical preference information and make the filled behavior more in line with the user's actual needs.

[0049] Suppose the set of similar products that the user has browsed historically is , for each product , calculate its browsing probability ,The browsing probability can be calculated by the ratio of the historical browsing times of ,products to the total browsing times of similar products, which makes the ,selection of filling products more scientific and reasonable, and avoids the ,irrationality that may be caused by random selection.

[0050] According to the calculated browsing probability , select from the product collection Products, select The products are inserted into the interruption point in the order of browsing probability from high to low to form a new behavior sequence, which makes the filled behavior sequence closer to the actual situation, effectively solves the problem of behavior sequence interruption, keeps the behavior sequence coherent, and provides more complete and accurate data for subsequent model learning, which helps to improve the model's ability to predict and recommend user behavior.

[0051] The data preprocessing unit 2 uses a knowledge graph-based encoding method to encode product features and behavior features. The specific steps are as follows:

[0052] In the field of e-commerce, there are complex relationships between commodities, brands, categories and attributes. Constructing a knowledge graph can clearly represent these relationships in a graphical way. The knowledge graph can integrate information from many aspects and provide rich semantic information and association basis for the subsequent encoding of commodity characteristics and behavioral characteristics, which helps to understand and process data more comprehensively and accurately. It collects commodity data, brand information, category classification and commodity attribute data from e-commerce platforms, and uses commodities, brands, categories and attributes as nodes, and edges to represent the relationship between them. For example, the commodity node is connected to the brand node through the "brand association" edge, the commodity node is connected to the category node through the "category association" edge, and the commodity node is connected to the corresponding attribute node through the "attribute association" edge, thereby constructing a knowledge graph in the field of e-commerce. ,in is a collection of nodes, It is a set of edges. The knowledge graph can intuitively display the complex relationships in the e-commerce field, provide a structured information basis for the subsequent encoding process, and facilitate the mining of potential connections between data.

[0053] For a given product , in the knowledge graph Find the corresponding category node in , Brand Node And the key attribute node collection ,Based on the knowledge graph search node, it is possible to accurately locate the relevant feature information of the product, avoiding the problem of information loss or inaccuracy that may occur in traditional methods.,In order to convert the node information in the knowledge graph into a numerical form that can be processed by a computer, the node needs to be embedded.,By splicing the embedding vectors of the category, brand and attribute nodes, the various features of the product can be integrated into one vector, forming a coding vector of the product features, which is convenient for subsequent model processing.

[0054] Use the graph embedding algorithm to embed the nodes in the knowledge graph and get the embedding vector of each node. The embedding vector of , brand node The embedding vector of , attribute node The embedding vector of , the encoding vector of product features is the concatenation of these vectors, ,in It represents the vector concatenation operation. The generated product feature encoding vector contains rich product information, which helps the model better understand the characteristics of the product and improve the performance of the model.

[0055] Define a set of behavioral characteristics , for a behavior , using one-hot encoding to map it into a length of Vector , of which only elements are 1 and the rest are 0. One-hot encoding is simple and intuitive, can clearly represent different behavior categories, and is easy to merge with the product feature encoding vector, converting the behavior features into a computable vector form, providing a basis for subsequent feature fusion.

[0056] For each behavior node, the one-hot encoding vector of the behavior The product feature encoding vector of the behavior node Fusion is performed to obtain the final encoding vector containing product and behavior information ,Feature fusion integrates the information of products and behaviors, so that the encoding vector can more comprehensively describe the interaction between users and products, providing richer input features for the model.

[0057] Model building training unit 3 introduces the long short-term memory network as the core model, takes the encoding vector of the user behavior sequence obtained after preprocessing as input, extracts and learns features through multi-layer long short-term memory network units, outputs the hidden state of each time step, and applies the attention mechanism on the hidden state. The attention weight of each time step is calculated, and the final user interest representation vector is obtained by weighted summation. Finally, the state, action and reward are defined, and the deep Q network is used as the reinforcement learning algorithm. The optimal recommended action is selected according to the current state, and the model parameters are updated according to the reward signal.

[0058] The model building training unit 3 includes an extraction learning module 31. When the multi-layer long short-term memory network unit in the extraction learning module 31 performs feature extraction and learning, a gating mechanism is introduced. The specific operations are as follows:

[0059] When processing complex sequence data, the input gate, forget gate, and output gate of the traditional long short-term memory network may not be able to dynamically and flexibly adjust the focus of feature extraction according to the input data. The introduction of adaptive adjustment gates can more finely screen and adjust information according to the current input and the previous state, thereby better capturing the long-term dependencies in the user behavior sequence and improving the feature extraction effect.

[0060] On the basis of the traditional long short-term memory network, an adaptive adjustment gate is added. For each time step in the input sequence, the input vector and the hidden state of the previous moment are set. The outputs of the input gate, forget gate and output gate are calculated through linear transformation and sigmoid function. Among them, the output gate is used to control how much new information can enter the cell state, the forget gate is used to control how much information of the cell state at the previous moment needs to be forgotten, and the output gate controls how much information of the cell state is output to the current hidden state. Finally, the output of the adaptive adjustment gate is calculated through linear transformation and hyperbolic tangent function. The adaptive adjustment gate dynamically adjusts the focus of feature extraction according to the current input and the previous state. The adaptive adjustment gate increases the flexibility and adaptability of the model, and can dynamically adjust the information processing method according to different input conditions, making the model more efficient when processing complex sequence data.

[0061] After the introduction of the adaptive adjustment gate, the hidden state at the current moment needs to be recalculated to make full use of the information filtered and adjusted by the adaptive adjustment gate, so as to more accurately represent the feature information at the current moment and provide better input for subsequent attention mechanisms and reinforcement learning.

[0062] First, the forget gate determines how much information of the hidden state at the previous moment is retained, and the input gate and the adaptive adjustment gate jointly determine how much new information enters the cell state. The two are added together and mapped to the hyperbolic tangent function The new hidden state calculation method comprehensively considers the role of the adaptive adjustment gate, which can more effectively integrate historical information and current input information, improve the representation ability of the hidden state, and enable the model to more accurately represent the user's behavioral characteristics at each time step.

[0063] The model building training unit 3 includes an attention mechanism module 32. When the attention mechanism module 32 applies the attention mechanism on the hidden state, a multi-head attention mechanism is adopted. The specific calculation method is as follows:

[0064] The traditional single attention mechanism may not be able to fully capture the different feature information in the hidden state sequence. Mapping the hidden state vector to multiple subspaces through linear transformation and generating multiple-head query vectors, key vectors, and value vectors can enable the model to pay attention to the hidden state sequence from different angles and subspaces, thereby extracting feature information more comprehensively and meticulously.

[0065] Let the hidden state sequence be ,in It is The hidden state vector of time steps, setting the number of heads to , for each head , generate the query vector through linear transformation , key vector Sum value vector ,The multi-head attention mechanism increases the expressive ability of the model and the diversity of ,feature extraction by mapping multiple subspaces, and can capture richer feature information in the ,hidden state sequence.

[0066] For each head , calculate the attention score , the formula is ;in It is The first The query vector for time steps, It is The first The key vector for each time step, is the dimension of the key vector, divided by This is to prevent the dot product result from being too large, causing the softmax function gradient to disappear, and then the attention weight is obtained through the softmax function ,The calculation of attention scores and weights can adaptively determine the importance of each time step in the current head, allowing the model to focus on more critical parts of the sequence, improving the efficiency and accuracy of information processing.

[0067] For each head , calculate the context vector , the formula is ;in It is The first The context vector integrates the information of the attention weight and the value vector, which can effectively capture the key information in the hidden state sequence, and the context vector of each head reflects the characteristics of the sequence from different subspaces and angles.

[0068] Will The context vector of each head Perform splicing to obtain the spliced ​​vector , and then through the linear transformation Get the final attention output , the formula is ;in It is the output weight matrix. The concatenation and linear transformation operations can fuse and integrate the feature information extracted by multiple heads to obtain a comprehensive and more representative feature representation, giving full play to the advantages of the multi-head attention mechanism. The final attention output can more comprehensively and accurately represent the user's interests, and provide better feature input for subsequent reinforcement learning and product recommendations, thereby improving the recommendation accuracy and effect of the e-commerce intelligent recommendation system.

[0069] The model building training unit 3 includes a reinforcement learning module 33. When the reinforcement learning module 33 uses a deep Q network as a reinforcement learning algorithm, the specific method of defining states, actions and rewards is as follows:

[0070] In reinforcement learning, accurately describing the state of the current environment is crucial for the intelligent agent to make reasonable decisions. Taking the current user's interest representation vector, the user's current e-commerce page information, and the user's historical behavior statistical characteristics as a state can comprehensively and meticulously characterize the user's current situation, provide rich contextual information for the recommendation system, and enable the intelligent agent to make recommendations that are more in line with user needs based on this state.

[0071] Assume that the current user's interest representation vector is , which is a vector that can reflect the user's interest obtained through the previous attention mechanism module 32. The information of the e-commerce page where the user is currently located can be represented by a vector Indicates that the user's historical behavior statistical features are represented by vector Indicates that it can contain statistical information about the number of items browsed, the number of purchases, and the average purchase amount by users in the past period of time. It can be expressed as , which enables the recommendation system to make more accurate product recommendations based on the user’s specific status, improving the relevance and accuracy of the recommendations, thereby enhancing the user’s shopping experience on the e-commerce platform.

[0072] Set up product library , select from Products form a candidate product set , each candidate product set Corresponding to a recommended action , which enables the recommendation system to provide users with richer and more personalized product recommendations, increasing the chances of users discovering products of interest, thereby improving users' purchase conversion rates and platform sales.

[0073] The reward mechanism is the core part of reinforcement learning. It is used to motivate the agent to take actions that can maximize long-term benefits. Different rewards are given according to the different feedback from users on the recommended products. It can guide the agent to learn which recommended actions are effective and which are ineffective, thereby continuously optimizing the recommendation strategy.

[0074] Let the positive reward be If the user sees the recommended product within a certain period of time Positive rewards will be given to those who have made purchases within If the user clicks on the recommended product and enters the details page, a half positive reward will be given If the user does not respond to the recommended product, a negative reward will be given. This reward setting can reasonably measure users' different feedback on recommended products, motivate the agent to give priority to recommended actions that can prompt users to buy or click, and avoid making invalid recommendations, thereby improving the efficiency and effectiveness of the recommendation system.

[0075] Guided by the reward mechanism, the intelligent agent can continuously adjust the recommendation strategy and improve the quality of recommendations, thereby increasing users' willingness to purchase and the platform's revenue, and achieving optimization and improvement of the recommendation system.

[0076] The recommendation generation output unit 4 selects products related to the user's interests from the product library according to the user's interest representation vector, forms a candidate product set, outputs a recommendation strategy, sorts the candidate product set, determines the order of recommended products, and finally displays the sorted recommended product list to the user, while recording the user's feedback on the recommendation results.

[0077] When ranking the candidate product set, the recommendation generation output unit 4 adopts a ranking algorithm that integrates user preferences and product popularity. The specific steps are as follows:

[0078] Users on e-commerce platforms often have different degrees of preference for products of different categories and brands. Understanding users' preferences for the categories and brands of candidate products can more accurately grasp users' interests. By obtaining user preference scores through weighted summation, we can comprehensively consider the impact of category and brand preferences on users' product selection, thereby providing a basis for product sorting that is more in line with user needs.

[0079] Let the candidate product be , which belongs to the category , the brand is , extract the user's preference for the category from the user behavior data and preference for the brand , assign weights to category preference and brand preference respectively and , then the user's preference for candidate products Preference score Calculation formula , taking into account users' preferences for categories and brands, it can more carefully characterize users' personalized needs, avoid the limitation of evaluating user preferences from only a single dimension, and make product sorting more in line with users' actual interests.

[0080] The popularity of a product reflects its popularity in the market. At a certain time The number of views in , the purchase amount is , assign preset weights to page views and purchases and , then the product The popularity score The calculation formula is ;in and are all candidate products at time The maximum number of views and the maximum number of purchases within a product are combined to measure the popularity, which can fully reflect the popularity of the product. The influence of the two on the popularity score can be flexibly adjusted through weight distribution, making the sorting result more reasonable.

[0081] Assign weights to user preference scores and product popularity scores respectively and , then the candidate product The ranking score The calculation formula is ,in It can be adjusted according to user groups, integrating two important factors: user preference and product popularity, to achieve an organic combination of personalized recommendations and popular recommendations. It can meet the needs of different users in different scenarios and improve the performance and adaptability of the recommendation system.

[0082] Collect user feedback on recommendation results, including purchase behavior, click behavior, dwell time, and skip behavior information. Organize this information according to certain rules to form structured data, use association rule mining algorithm to analyze structured feedback data, mine association rules that meet the conditions, and feed the analysis results back to each unit of the recommendation system, thus realizing closed-loop optimization of the recommendation system. It can continuously adjust and improve the recommendation strategy according to the actual feedback of users, improve the adaptability and accuracy of the recommendation system, and enable the recommendation system to continuously meet the needs of users.

[0083] In the present invention, a data collection and storage unit 1 is used to deploy embedding points on each page of the e-commerce platform to collect data such as user behavior, equipment, geographic location and access time and store them in a distributed database. A data preprocessing unit 2 extracts data, constructs a user behavior sequence, and encodes the product and behavior characteristics. A model construction and training unit 3 uses a long short-term memory network in combination with an attention mechanism to obtain user interest vectors, and updates model parameters with the help of a deep Q network. A recommendation generation and output unit 4 filters, sorts and displays products based on the interest vectors, and collects feedback to optimize the system, thereby improving recommendation accuracy, enhancing user shopping experience, and helping e-commerce platforms improve operational efficiency and product sales conversion rates.

[0084] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. E-commerce intelligent recommendation system based on user behavior big data, characterized by: It includes a data acquisition and storage unit (1), a data preprocessing unit (2), a model building and training unit (3), and a recommendation generation and output unit (4); The data collection and storage unit (1) deploys tracking points on each page of the e-commerce platform, wherein the tracking points are used to collect user behavior data and user device information, geographic location and access time, and store the collected behavior data in a distributed database; The data preprocessing unit (2) extracts and processes the collected behavior data from the distributed database, and then arranges the user behavior data in chronological order to construct a user behavior sequence, wherein each sequence contains a plurality of behavior nodes, and each node contains a product feature and a behavior feature, and finally encodes the product feature and the behavior feature and converts them into a numerical vector; The model construction training unit (3) introduces a long short-term memory network as a core model, takes the encoding vector of the user behavior sequence obtained after preprocessing as input, extracts and learns features through multi-layer long short-term memory network units, outputs the hidden state of each time step, applies an attention mechanism to the hidden state, calculates the attention weight of each time step, obtains the final user interest representation vector by weighted summation, and finally defines the state, action and reward, adopts a deep Q network as a reinforcement learning algorithm, selects the optimal recommended action according to the current state, and updates the model parameters according to the reward signal; The recommendation generation output unit (4) selects products related to the user's interests from the product library according to the user's interest representation vector, forms a candidate product set, outputs a recommendation strategy, sorts the candidate product set, determines the order of recommended products, and finally displays the sorted recommended product list to the user, while recording the user's feedback on the recommendation results.

2. The e-commerce intelligent recommendation system based on user behavior big data according to claim 1 is characterized by: In the data collection and storage unit (1), an adaptive sampling strategy is adopted when collecting user behavior data. The specific steps are as follows: Define a set of behavior data importance indicators, where the behavior data importance indicators are the correlation between behavior and purchase conversion, the rarity of behavior, and the timeliness of behavior, and assign an initial weight to each indicator; For each user behavior event, its importance score is calculated according to the behavior data importance index and its initial weight, and a dynamic sampling rate is set according to the importance score. Adaptive sampling is performed through the dynamic sampling rate to obtain user behavior data.

3. The e-commerce intelligent recommendation system based on user behavior big data according to claim 2 is characterized by: When constructing a user behavior sequence, the data preprocessing unit (2) specifically handles the problem of behavior sequence interruption caused by long-term inactivity as follows: A time interval threshold is set. When the time interval between two adjacent behavior nodes in the user behavior sequence is found to be greater than the time interval threshold, the sequence is determined to be interrupted. For interrupted sequences, a sequence filling strategy is adopted, that is, analyzing the behavioral characteristics before the interruption. If the previous behavior is concentrated on browsing products in a certain category, several similar products are selected from the user's historical browsing history and inserted into the interruption according to the browsing probability to simulate the user's subsequent behavior.

4. The e-commerce intelligent recommendation system based on user behavior big data according to claim 3 is characterized by: The data preprocessing unit (2) adopts a knowledge graph-based encoding method when encoding the product features and behavior features. The specific steps are as follows: Construct a knowledge graph in the e-commerce field. The nodes include products, brands, categories, and attributes. The edges represent the relationships between them. For product features, find the corresponding category, brand, and key attribute nodes based on the knowledge graph, and concatenate the embedding vectors of these nodes as the encoding vector of the product features. For behavioral features, one-hot encoding is used to map them into corresponding behavior vectors, which are then fused with the product feature encoding vector of the behavior node to obtain the final encoding vector containing product and behavior information.

5. The e-commerce intelligent recommendation system based on user behavior big data according to claim 4 is characterized by: The model building training unit (3) includes an extraction learning module (31). The multi-layer long short-term memory network unit in the extraction learning module (31) introduces a gating mechanism when performing feature extraction and learning. The specific operation is as follows: Based on the input gate, forget gate and output gate of the long short-term memory network, an adaptive adjustment gate is added. For each time step in the input sequence, the input vector and the hidden state of the previous moment are set, and the outputs of the input gate, forget gate and output gate are calculated through linear transformation and sigmoid function. Among them, the output gate is used to control how much new information can enter the cell state, the forget gate is used to control how much information of the cell state at the previous moment needs to be forgotten, and the output gate controls how much information of the cell state is output to the current hidden state. Finally, the output of the adaptive adjustment gate is calculated through linear transformation and hyperbolic tangent function. The adaptive adjustment gate dynamically adjusts the feature extraction according to the current input and the previous state.

6. The e-commerce intelligent recommendation system based on user behavior big data according to claim 5 is characterized by: The model building training unit (3) includes an attention mechanism module (32). When the attention mechanism module (32) applies the attention mechanism to the hidden state, a multi-head attention mechanism is adopted. The specific calculation method is as follows: Set the hidden state sequence, extract the hidden state vector from the hidden state sequence, and map the hidden state vector to multiple subspaces through linear transformation to generate The query vector, key vector, and value vector of each head; For each head, calculate the attention score, and get the attention weight through the softmax function, then calculate the context vector of each head, and finally The context vectors of the heads are concatenated and linearly transformed to obtain the final attention output.

7. The e-commerce intelligent recommendation system based on user behavior big data according to claim 6 is characterized by: The model building training unit (3) includes a reinforcement learning module (33). When the reinforcement learning module (33) uses a deep Q network as a reinforcement learning algorithm, the specific method of defining states, actions and rewards is as follows: The state consists of the current user's interest representation vector, the user's current e-commerce page information, and the user's historical behavior statistical characteristics; Select a certain number of products from the product library as recommended candidate actions, and each action corresponds to a set of candidate products; If the user purchases the recommended product within a certain period of time after seeing it, a positive reward will be given. If the user clicks on the recommended product to enter the details page, a half positive reward will be given. If the user does not respond to the recommended product, a negative reward will be given.

8. The e-commerce intelligent recommendation system based on user behavior big data according to claim 7 is characterized by: The recommendation generation output unit (4) uses a sorting algorithm that integrates user preferences and product popularity when sorting the candidate product set. The specific steps are as follows: For each candidate product, calculate its user preference score, extract the user's preference for the product category and brand from the user behavior data, and obtain the user preference score through weighted summation; Calculate the popularity score of the product by counting the number of views and purchases of the product within a certain period of time, summarizing them according to the preset weights, and finally calculating the ranking score of the candidate products based on the user preference score and the product popularity score; Collect user feedback on recommendation results, organize the feedback data into structured data form, use association rule mining algorithm to analyze the feedback data, and optimize each unit of the recommendation system based on the analysis results.

Images