E-commerce user behavior analysis method based on deep learning

Abstract

The invention relates to the technical field of e-commerce, and discloses an e-commerce user behavior analysis method based on deep learning, which comprises the following steps: acquiring a historical browsing record and a historical purchase record of a current user; obtaining multi-dimensional attribute information of the commodity, wherein the multi-dimensional attribute information comprises price, brand, evaluation score and sales volume; recording behavior data of the user when the user browses the commodity recommendation list in real time, wherein the behavior data comprises click times and browsing duration; obtaining timeliness information of the commodity, wherein the timeliness information comprises a promotion activity and an inventory state; according to the historical browsing record and the historical purchase record, calculating preference scores of the current user for the type information of the multiple commodities; constructing a collaborative analysis model based on the preference score, the commodity multi-dimensional attribute information, the user behavior data and the commodity timeliness information; according to the collaborative analysis model, generating recommendation items and a display sequence in a recommendation list; and displaying the recommended items on a display interface according to the display sequence.

Description

Technical Field

[0001] The present invention relates to the technical field of e-commerce, and in particular to an e-commerce user behavior analysis method based on deep learning. Background Art

[0002] With the rapid development of e-commerce, user behavior analysis methods based on data mining have become a key technology to improve user experience and the competitiveness of e-commerce platforms. By analyzing users' historical browsing records and historical purchase records, e-commerce platforms can identify users' interests and preferences, thereby generating personalized product recommendation lists. Such recommendation systems can help users quickly find products of interest, improve the convenience of using the platform, and promote product sales.

[0003] Existing e-commerce user behavior analysis methods based on data mining (such as Chinese patent CN118134608A) calculate the user's preference scores for different types of goods through the user's historical behavior data, and record the user's behavior when browsing the product recommendation list, such as the first recommended item, exit time, etc. Based on this data, the system can provide the user with product recommendations related to the preference when the user opens the recommendation list again, so that the user can quickly return to the product that he was interested in before, improving the user experience.

[0004] However, existing technologies mainly rely on the user's historical preference score when optimizing the display order of the recommendation list, and do not fully consider the impact of the multi-dimensional attributes of the product on the user's decision. For example, when users choose a product, they may be affected by multiple factors such as product price, brand, evaluation score, sales volume, etc. A single preference score may not fully reflect the user's true interest in the product, thus limiting the accuracy and effectiveness of the recommendation system.

[0005] In addition, the user's behavioral data when browsing the recommendation list (such as the number of clicks, browsing time) and the timeliness information of the product itself (such as promotions, inventory status) may also have a significant impact on the recommendation results. However, existing methods fail to effectively integrate these dynamic factors for collaborative analysis. This deficiency may cause the order of items in the recommendation list to be inconsistent with the user's actual needs, thereby reducing user experience and platform revenue.

[0006] In summary, the existing user behavior analysis methods based on data mining still have room for improvement in optimizing the display order of recommendation lists. In order to solve this problem, it is necessary to provide a new technical solution that combines the user's preference score with the multi-dimensional attribute information of the product and establishes a collaborative analysis model to improve the accuracy of the recommendation list and user satisfaction. Summary of the invention

[0007] In view of the deficiencies in the prior art, the present invention provides an e-commerce user behavior analysis method based on deep learning, which combines the user's preference scores with the multi-dimensional attribute information of the goods and establishes a collaborative analysis model to improve the accuracy of the recommendation list and user satisfaction.

[0008] In order to achieve the above object, the present invention adopts the following technical solutions: A method for analyzing e-commerce user behavior based on deep learning, comprising: Get the current user's browsing history and purchase history; Obtain multi-dimensional attribute information of products, including price, brand, evaluation score and sales volume; Real-time recording of user behavior data when browsing product recommendation lists, including click counts and browsing time; Get product timeliness information, including promotions and inventory status; Calculate the current user's preference scores for the type information of multiple commodities based on historical browsing records and historical purchase records; Build a collaborative analysis model based on preference scores, multi-dimensional product attribute information, user behavior data, and product timeliness information; Generate recommended items and display order in a recommendation list according to the collaborative analysis model; On the display interface, the recommended items are displayed according to the display order.

[0009] Preferably, it includes: capturing the user's behavioral characteristics in the current browsing session in real time, including the number of clicks, browsing time and sliding behavior, and calculating the user's short-term interest score according to the percentage of clicks, browsing time and sliding speed, as follows: , where C, D, and M are the percentage of clicks, percentage of browsing time, and sliding behavior preference scores, respectively. , , is the weight parameter, satisfying + + =.

[0010] Preferably, the comprehensive score Stotal is dynamically calculated by combining the short-term interest score Sshort, the historical interest score Shistory and the product timeliness score St, and the formula is: ,in , , is the weight parameter, satisfying + + s=.

[0011] Preferably, the user logs in to the e-commerce platform through a unique account information, and the system identifies and binds multiple devices according to the user account, and records the browsing data on different devices; Collect users’ click behavior, browsing time and sliding behavior in real time on multiple devices, and integrate these behavior data through user accounts; The product preference score on each device is calculated by taking the weighted average of click behavior, browsing time, and sliding behavior, and a comprehensive preference score is generated based on the device weight. The display order of the product recommendation list is adjusted based on the comprehensive preference score.

[0012] Preferably, ; in , , The product preference scores on mobile phones, tablets and desktop devices are respectively, is the device weight, satisfying =, the integration of cross-device behavior data includes the weighted average of the user's click behavior, browsing time and sliding behavior on multiple devices.

[0013] Preferably, including: Build a user preference model, analyze the user's historical browsing history, purchase history and click behavior, and obtain the user's comprehensive preference score for the product; Build a product attribute scoring model to calculate the comprehensive attribute score of the product based on its price, brand, evaluation score and other attributes; The final recommendation score is generated by weighted average based on the user's preference score and the product's attribute score; Adjust the display order of the product recommendation list based on the final recommendation score.

[0014] Preferably, the user preference score calculation formula is: ; in Provide users with product attributes The preference value, is the weight of the corresponding attribute.

[0015] Preferably, the product attribute score calculation formula is:

[0016] in , , They are the price, brand, evaluation score and sales score of the product. and is the weight of the corresponding attribute.

[0017] Preferably, the final recommendation score calculation formula is: ; in Rate user preferences, Rate product attributes. The weight for balancing user preferences and product attribute ratings.

[0018] An e-commerce user behavior analysis system based on deep learning, comprising: Data collection module: used to obtain the current user's historical browsing history, historical purchase history, and multi-dimensional attribute information of the product, including price, brand, evaluation score, and sales volume. At the same time, it collects the user's behavior data when browsing the recommendation list in real time, including the number of clicks and browsing time. Product information collection module: used to obtain the timeliness information of products, including promotional activities and inventory status; User preference calculation module: Calculates the user's preference scores for various types of product information based on historical browsing records and historical purchase records; Collaborative analysis module: Based on the user's preference score, multi-dimensional attribute information of the product, user behavior data and product timeliness information, a collaborative analysis model is constructed to determine the priority and display order of the recommended items in the recommendation list; Recommendation generation module: generates an optimized recommendation list based on the output of the collaborative analysis module; Display module: Display the recommended items on the display interface in the order in which they are displayed in the recommendation list; Feedback optimization module: records user feedback on the recommendation list, and updates the collaborative analysis model based on the feedback data to continuously optimize the recommendation effect.

[0019] Compared with the prior art, the present invention has the following beneficial effects: 1. Accurate recommendations: By comprehensively considering multi-dimensional user data and product information, we can more accurately analyze user needs, generate recommendation lists that are more in line with user interests, increase the probability of users finding their desired products, and thus improve user satisfaction.

[0020] 2. Real-time adaptation: Calculating short-term interest scores and comprehensive scores can timely capture the user's current interest changes, and dynamically adjust the recommendation list based on these changes, so that the recommendation results are more in line with the user's current needs.

[0021] 3. Cross-device optimization: Integrate user behavior data on different devices to generate comprehensive preference scores across devices, solve the problem of inconsistent recommendations in multi-device usage scenarios, and provide users with a seamless shopping experience.

[0022] 4. Comprehensive consideration of user and product attributes: By building a user preference model and a product attribute scoring model, and combining the two to generate the final recommendation score, the impact of user preferences and the product's own attributes on the recommendation is fully considered, further optimizing the ranking of the recommendation list. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0024] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0025] The present invention is further described in detail below with reference to the accompanying drawings.

[0026] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles of the present invention defined in the following description can be used for other embodiments, variations, improvements, equivalents, and other technical solutions that do not deviate from the spirit and scope of the present invention.

[0027] Those skilled in the art should understand that, in the disclosure of the present invention, the directions or positions indicated by the terms "longitudinal", "lateral", "up", "down", "left", "right", "front", "back", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc. are based on the directions or positional relationships shown in the accompanying drawings, which are only simplified descriptions for the convenience of describing the present invention, and do not indicate or imply that the device or component referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, the above terms should not be understood as limitations on the present invention.

[0028] It is to be understood that the term "one" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element may be one, while in another embodiment, the number of the element may be multiple, and the term "one" should not be understood as a limitation on the quantity.

[0029] See also Figure 1, an e-commerce user behavior analysis method based on deep learning, comprising: 1. Data acquisition: Get the current user's browsing history and purchase history; Obtain multi-dimensional attribute information of products, including price, brand, evaluation score and sales volume; Real-time recording of user behavior data when browsing product recommendation lists, including click counts and browsing time; Get product timeliness information, including promotions and inventory status.

[0030] 2. Preference score calculation: Based on historical browsing records and historical purchase records, calculate the current user's preference scores for the type information of multiple commodities.

[0031] 3. Collaborative analysis model construction: A collaborative analysis model is constructed based on preference scores, multi-dimensional attribute information of products, user behavior data and product timeliness information.

[0032] 4. Recommendation list production and reality: Generate recommended items and display order in a recommendation list according to the collaborative analysis model; On the display interface, the recommended items are displayed according to the display order.

[0033] 5. Short-term interest score calculation: Capture user behavior characteristics in real time during the current browsing session, including clicks, browsing time, and sliding behavior.

[0034] Calculate the user's short-term interest score based on the percentage of clicks, browsing time and sliding speed, and define the short-term interest score , the formula is: , in: : The percentage of times users click on a certain type of product. The value range is [0, 1] and is calculated as: , : The number of times users click on this type of product; : Total number of user clicks.

[0035] : User browsing time ratio, the value range is [0, 1], calculated as: , : The browsing time of users on this category of products; : Total browsing time.

[0036] : The preference score of the user's sliding behavior, with a value range of [0, 1]. The slower the sliding speed (i.e., the longer the product stays), the higher the score. It is calculated as: , : User sliding speed; : Maximum sliding speed.

[0037] , , is a weight parameter used to control the impact of clicks, browsing time, and sliding behavior, satisfying + + =1.

[0038] 6. Calculation of comprehensive score: Combined with short-term interest scores , Historical Interest Rating and product timeliness rating The comprehensive score is calculated using the formula: , in , , are the weights of historical interest, timeliness and short-term interest, respectively, satisfying + + s=1, according to Dynamically adjust the recommendation list to ensure that the recommendation order is consistent with the user's current needs.

[0039] The comprehensive score calculation example: Assume that a user prefers electronic products in his historical behavior = 0.8, the current session mainly browses home furnishings, and the short-term interest score captured in real time is =0.9, and the product timeliness score is =0.6. The weight is set to =0.4, =0.3, =0.3, the comprehensive score is calculated as follows: =0.4×0.8+0.3×0.6+0.3×0.9=0.77, The recommendation system adjusts the priority display order of household items based on the comprehensive scores to better meet the users' short-term needs.

[0040] 7. Cross-device data processing: Device identification and account binding When a user logs in to an e-commerce platform, the system binds different devices (such as mobile phones, tablets, computers, etc.) through unique account information (such as mobile phone number, email address, social media account). Every time a user logs in on a different device, the system records the device information and associates it with the user account.

[0041] Cross-device behavioral data collection When users browse products, the system collects browsing data on each device in real time, including click behavior, browsing time, sliding behavior, etc., and uniformly identifies this data through the user account.

[0042] Cross-device behavior matching and integration The system integrates the behavior data of multiple devices into the complete behavior data of a user in the following ways: Click behavior matching: If a user clicks the same product on different devices, the click behavior is considered a duplicate record of the same behavior.

[0043] Browsing time matching: The browsing time of the same product on multiple devices is weighted averaged to calculate the total browsing time of the product.

[0044] Sliding behavior integration: Based on the persistence of sliding behavior (such as sliding dwell time), sliding behaviors across devices are integrated into a unified data model through weighted averaging.

[0045] Cross-device behavior analysis model The system calculates the cross-device preference score for each product based on the integrated cross-device data and the browsing patterns of users on each device (for example, mobile devices usually browse quickly, while desktop devices browse for a longer time). The system adjusts the display order of the product recommendation list based on the comprehensive preference score. The scoring formula is: , in: , , The product preference scores on mobile phones, tablets and desktop devices are respectively, is the device weight, satisfying =1, respectively controlling the impact of different devices on the final preference score.

[0046] Cross-device behavior impacts recommended list Through integrated cross-device preference scoring Update the recommended list of items and dynamically adjust the display order to give the most relevant recommendations based on user behavior on different devices.

[0047] The cross-device data processing embodiment thereof: Assume that user A browsed product X on a mobile phone and clicked on it, staying on it for 20 seconds, browsed product X again on a tablet and stayed on it for 40 seconds, and viewed product X on a desktop device but did not click on it, staying on it for 10 seconds. The cross-device behavior data can be integrated in the following ways: The rating on mobile is =0.8 (clicking behavior results in a higher score).

[0048] The rating on the tablet is t=0.9 (longer browsing time leads to higher ratings).

[0049] The rating on desktop is =0.6 (no clicks, but shorter browsing time).

[0050] Assume that the device weight is set to =0.4 =0.3, =0.3, the final score across devices is: =0.4×0.8+0.3×0.9+0.3×0.6=0.78, The comprehensive score will affect the display order of product X in the recommendation list, thereby better meeting the user's cross-device preferences.

[0051] 8. Construction of multi-dimensional scoring model: 1) Construction of user preference model First, we build a user preference model by analyzing the user's historical browsing history, purchase history, click behavior and other data. We assume that the user's preference for products is mainly affected by the following factors: Product type (e.g., electronics, clothing, etc.); Product prices; Product brand; Product review score; The calculation formula of the user purchase history preference model is: , in: Score the comprehensive preference of user u for the product; For user u The preference value of each product attribute (such as price, brand, etc.); is the weight of each product attribute, indicating the degree of influence of the attribute on user preference; n is the number of product attributes considered.

[0052] 2) Construction of product attribute scoring model The multi-dimensional attributes of the product itself (price, brand, evaluation score, etc.) will also affect the recommendation ranking. The attribute scoring model of the product can be calculated using the following formula: , in: Score the comprehensive attributes of product g; , , They are the price, brand, evaluation and sales ratings of the product; and is the weight of the corresponding attribute.

[0053] 3) Comprehensive rating of user preferences and product attributes The user's preference score is combined with the product attribute score to obtain the final recommendation ranking score. The final recommendation score Rg can be calculated by the following formula: , in Rate user preferences, Rate product attributes. For goods The final recommendation score, Score the user's preference for the product. Rate the attributes of the product. To balance the weight of user preference and product attribute rating, usually 0≤ ≤1.

[0054] 4) Recommendation list generation and sorting The system generates and optimizes a list of recommended products, sorts them from high to low according to their ratings, and finally displays them to users based on the final recommended rating Rg of each product.

[0055] The multi-dimensional scoring model construction example: Assume that user A's preference for product X is as follows: Product X is priced relatively low and has a rating of 0.8; Product X brand is very important to user A, with a rating of 0.9; Product X has a review score of 4.5 and a rating of 0.85; Product X has a high sales volume and a rating of 0.7.

[0056] Assume that the preference score of product X is =0.85, the weights of price, brand, evaluation score, and sales volume are =0.2, =0.3, =0.3, =0.2, attribute score of product X Calculated as: =0.2×0.8+0.3×0.9+0.3×0.85+0.2×0.7=0.82 Then, combining user preferences and product attributes, the final recommendation score is calculated using the following formula: Assumptions =0.6, the final score Rg is calculated as: Rg=0.6×0.85+0.4×0.82=0.845 According to the final score Rg, the recommendation order of product X will be sorted in the recommendation list according to its score, and products with high scores will be displayed to users first.

[0057] The system architecture is: Data collection module: used to obtain the current user's historical browsing history, historical purchase history, and multi-dimensional attribute information of the product, including price, brand, evaluation score, and sales volume. At the same time, it collects the user's behavior data when browsing the recommendation list in real time, including the number of clicks and browsing time. Product information collection module: used to obtain the timeliness information of products, including promotional activities and inventory status; User preference calculation module: Calculates the user's preference scores for various types of product information based on historical browsing records and historical purchase records; Collaborative analysis module: Based on the user's preference score, multi-dimensional attribute information of the product, user behavior data and product timeliness information, a collaborative analysis model is constructed to determine the priority and display order of the recommended items in the recommendation list; Recommendation generation module: generates an optimized recommendation list based on the output of the collaborative analysis module; Display module: Display the recommended items on the display interface in the order in which they are displayed in the recommendation list; Feedback optimization module: records user feedback on the recommendation list, and updates the collaborative analysis model based on the feedback data to continuously optimize the recommendation effect.

[0058] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and the embodiments of the present invention may be deformed or modified in any way without departing from the principles.

Claims

1. A method for analyzing e-commerce user behavior based on deep learning, characterized in that: include: Get the current user's browsing history and purchase history; Obtain multi-dimensional attribute information of products, including price, brand, evaluation score and sales volume; Real-time recording of user behavior data when browsing product recommendation lists, including click counts and browsing time; Get product timeliness information, including promotions and inventory status; Calculate the current user's preference scores for the type information of multiple commodities based on historical browsing records and historical purchase records; Build a collaborative analysis model based on preference scores, multi-dimensional product attribute information, user behavior data, and product timeliness information; Generate recommended items and display order in a recommendation list according to the collaborative analysis model; On the display interface, the recommended items are displayed according to the display order.

2. The e-commerce user behavior analysis method based on deep learning according to claim 1 is characterized in that: include: Capture the user's behavioral characteristics in the current browsing session in real time, including the number of clicks, browsing time and sliding behavior, and calculate the user's short-term interest score based on the percentage of clicks, browsing time and sliding speed, as follows: , where C, D, and M are the percentage of clicks, percentage of browsing time, and sliding behavior preference scores, respectively. , , is the weight parameter, satisfying + + =1.

3. The e-commerce user behavior analysis method based on deep learning according to claim 2 is characterized in that: By combining the short-term interest score Sshort, the historical interest score Shistory and the product timeliness score St, the comprehensive score Stotal is dynamically calculated. The formula is: ,in , , is the weight parameter, satisfying + + s=1.

4. The e-commerce user behavior analysis method based on deep learning according to claim 1 is characterized in that: Users log in to the e-commerce platform with their unique account information. The system identifies and binds multiple devices based on the user account and records browsing data on different devices. Collect users’ click behavior, browsing time and sliding behavior in real time on multiple devices, and integrate these behavior data through user accounts; The product preference score on each device is calculated by taking the weighted average of click behavior, browsing time, and sliding behavior, and a comprehensive preference score is generated based on the device weight. The display order of the product recommendation list is adjusted based on the comprehensive preference score.

5. The method for analyzing e-commerce user behavior based on deep learning according to claim 4 is characterized in that: ； in , , The product preference scores on mobile phones, tablets and desktop devices are respectively, is the device weight, satisfying =1, the integration of cross-device behavior data includes the weighted average of the user's click behavior, browsing time and sliding behavior on multiple devices.

6. The e-commerce user behavior analysis method based on deep learning according to claim 1 is characterized in that: include: Build a user preference model, analyze the user's historical browsing history, purchase history and click behavior, and obtain the user's comprehensive preference score for the product; Build a product attribute scoring model to calculate the comprehensive attribute score of the product based on its price, brand, evaluation score and other attributes; The final recommendation score is generated by weighted average based on the user's preference score and the product's attribute score; Adjust the display order of the product recommendation list based on the final recommendation score.

7. The e-commerce user behavior analysis method based on deep learning according to claim 6 is characterized in that: The user preference score calculation formula is: ； in Provide users with product attributes The preference value, is the weight of the corresponding attribute.

8. The e-commerce user behavior analysis method based on deep learning according to claim 7 is characterized in that: The formula for calculating the product attribute score is: ； in , , They are the price, brand, evaluation score and sales score of the product. and is the weight of the corresponding attribute.

9. According to the e-commerce user behavior analysis method based on deep learning according to claim 8, the final recommendation score calculation formula is: ； in Rate user preferences, Rate product attributes. The weight for balancing user preferences and product attribute ratings.

10. An e-commerce user behavior analysis system based on deep learning, characterized in that: include: Data collection module: used to obtain the current user's historical browsing history, historical purchase history, and multi-dimensional attribute information of the product, including price, brand, evaluation score, and sales volume. At the same time, it collects the user's behavior data when browsing the recommendation list in real time, including the number of clicks and browsing time. Product information collection module: used to obtain the timeliness information of products, including promotional activities and inventory status; User preference calculation module: Calculates the user's preference scores for various types of product information based on historical browsing records and historical purchase records; Collaborative analysis module: Based on the user's preference score, multi-dimensional attribute information of the product, user behavior data and product timeliness information, a collaborative analysis model is constructed to determine the priority and display order of the recommended items in the recommendation list; Recommendation generation module: generates an optimized recommendation list based on the output of the collaborative analysis module; Display module: Display the recommended items on the display interface in the order in which they are displayed in the recommendation list; Feedback optimization module: records user feedback on the recommendation list, and updates the collaborative analysis model based on the feedback data to continuously optimize the recommendation effect.

Images