Commodity recommendation system and method for electronic shopping mall

Abstract

The invention discloses a commodity recommendation system and method for an electronic mall, and the system comprises a user multi-dimensional behavior insight module which is used for collecting and analyzing user behavior data, and outputting a user behavior feature vector; the commodity feature deep analysis module is used for acquiring and analyzing commodity information and outputting commodity feature vectors; the situation awareness recommendation engine module is used for acquiring a user behavior feature vector and a commodity feature vector, collecting current situation information and generating a commodity recommendation list; the recommendation result integration output module is used for receiving the commodity recommendation list, integrating and optimizing the commodity recommendation list and generating a recommendation display interface; and the recommendation optimization emergency decision module is used for acquiring and analyzing feedback information of all recommendation results and generating a recommendation optimization signal. Multiple modules are associated, multi-dimensional behavior data of users are collected and analyzed, potential features of commodities are deeply mined, a commodity recommendation list is generated in combination with real-time situation information, feedback of the users to recommendation results is collected, and system optimization is carried out in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent recommendation systems, in particular to a product recommendation system and method for an electronic mall. BACKGROUND

[0002] An electronic mall, also known as an e-commerce mall, is a virtual commercial space based on Internet technology that provides a trading platform for goods or services for merchants and consumers. It breaks the time and space limitations of traditional physical malls, allowing merchants and consumers to conduct transactions anytime and anywhere through electronic devices. Scientific product recommendations by the electronic mall are beneficial to improving its sales performance.

[0003] However, in the existing product recommendation system of an electronic mall, the common recommendation method is mainly based on the analysis of simple data such as user's historical purchase records and browsing records, and the recommendation algorithm is relatively single, which cannot fully consider the user's potential needs and interest changes, and lacks effective integration and analysis mechanism when dealing with multi-dimensional data, resulting in insufficient accuracy of the recommendation results. In addition, it cannot track feedback and reasonably analyze the recommendation performance and system optimization management status, which is not conducive to continuously improving the recommendation performance and effect. SUMMARY

[0004] In order to overcome the above-mentioned shortcomings of the prior art, the purpose of the present application is to provide a product recommendation system and method for an electronic mall.

[0005] The technical scheme adopted by the present application to solve its technical problems is: a product recommendation system for an electronic mall, comprising: a user multi-dimensional behavior insight module for collecting and analyzing user behavior data and outputting a user behavior feature vector; a product feature depth analysis module for obtaining and analyzing product information and outputting a product feature vector; a context-aware recommendation engine module for obtaining the user behavior feature vector and the product feature vector, collecting current context information, and generating a product recommendation list; a recommendation result integration output module for receiving the product recommendation list and integrating and optimizing it to generate a recommendation display interface; a recommendation optimization urgency decision module for obtaining feedback information of all recommendation results and analyzing it to generate a recommendation optimization signal.

[0006] As a further improvement of the present application: further comprising a dynamic tracking feedback module for collecting feedback information of the user on the recommendation results, and being communicatively connected to the recommendation optimization urgency decision module.

[0007] As a further improvement of the present application: the step of generating a recommendation optimization signal by the recommendation optimization urgency decision module comprises: All users involved in the commodity recommendation in the detection period are acquired, and marked as i, i is a natural number greater than 1; The purchase rate, the collection rate and the click rate of the user i for all the recommendation results in the detection period are collected, the corresponding weight coefficients are determined, and the recommendation feedback coefficients are calculated by weighted summation; If the recommendation feedback coefficient exceeds the preset recommendation feedback coefficient threshold, the user i is marked as a superior feedback object, otherwise, the user i is marked as an inferior feedback object; The inferior feedback summary value is calculated, which is the ratio of the number of inferior feedback objects in the detection period to the total number of users involved in the detection period; If the inferior feedback summary value exceeds the preset inferior feedback summary threshold, a recommendation optimization high emergency signal is generated.

[0008] As a further improvement of the application, the step that the recommendation optimization emergency decision module generates a recommendation optimization signal further comprises: If the inferior feedback summary value does not exceed the preset inferior feedback summary threshold, the average value of the recommendation feedback coefficients of all users is calculated to obtain a recommendation feedback table value; If the recommendation feedback table value does not exceed the preset recommendation feedback table threshold, a recommendation optimization high emergency signal is generated; If the recommendation feedback table value exceeds the preset recommendation feedback table threshold, a recommendation optimization low emergency signal is generated.

[0009] As a further improvement of the application, the recommendation optimization emergency decision module is communicatively connected to an optimization management evaluation module; The optimization management evaluation module is configured to acquire the number of recommendation optimization high emergency signals in an evaluation period and mark it as an optimization evaluation anomaly detection value; If the optimization evaluation anomaly detection value exceeds the preset optimization evaluation anomaly detection threshold, an optimization evaluation anomaly signal is generated; If the optimization evaluation anomaly detection value does not exceed the preset optimization evaluation anomaly detection threshold, the time when the recommendation optimization high emergency signal is generated is marked as an anomaly generation time, and the time when the optimization improvement measure is taken is marked as an optimization time.

[0010] As a further improvement of the application, the optimization management evaluation module is further configured to The optimization time and the anomaly generation time are calculated to obtain an optimization holding time value, all optimization holding time values in the evaluation period are acquired to calculate the average value and obtain an optimization time value; The number of optimization holding time values exceeding the preset optimization holding time threshold in the evaluation period is marked as an optimization performance value; In the interval length of adjacent recommendation optimization low emergency signals in the evaluation period, the number of generated recommendation optimization high emergency signals is marked as a high emergency duration value, and the maximum high emergency duration value in the evaluation period is marked as a high emergency holding amplitude value; The optimization management evaluation coefficient is calculated by weighted summation of the optimization evaluation abnormality detection value, the optimization situation value, the optimization performance value and the high emergency holding amplitude value.

[0011] As a further improvement of the present application, the optimization management evaluation module is further configured to If the optimization management evaluation coefficient exceeds the preset optimization management evaluation coefficient threshold, an optimization evaluation abnormality signal is generated. If the optimization management evaluation coefficient does not exceed the preset optimization management evaluation coefficient threshold, an optimization evaluation qualified signal is generated.

[0012] The present application also discloses a commodity recommendation method for an electronic mall, comprising: The user behavior data is analyzed to obtain a user behavior feature vector, the commodity information is analyzed to obtain a commodity feature vector, and a commodity recommendation list is generated in combination with the current context information; The feedback information of the user on the recommendation result is collected and a recommendation feedback coefficient is calculated, and the superior feedback object and the inferior feedback object are determined. The number proportion of the inferior feedback object is calculated to obtain an inferior feedback summary value, if the inferior feedback summary value exceeds the preset inferior feedback summary threshold, a recommendation optimization high emergency signal is generated, otherwise, the mean value of the recommendation feedback coefficient is calculated to obtain a recommendation feedback table value; if the recommendation feedback table value exceeds the preset recommendation feedback table threshold, a recommendation optimization low emergency signal is generated, otherwise, a recommendation optimization high emergency signal is generated.

[0013] As a further improvement of the present application, the present application further comprises: The number of times of generating the recommendation optimization high emergency signal in the evaluation period is obtained and is marked as an optimization evaluation abnormality detection value. If the optimization evaluation abnormality detection value exceeds the preset optimization evaluation abnormality detection threshold, an optimization evaluation abnormality signal is generated, otherwise, the time of generating the recommendation optimization high emergency signal is marked as an abnormality generation time, and the time of taking the optimization improvement measure is marked as an optimization time, and the time difference between the optimization time and the abnormality generation time is calculated to obtain an optimization holding time value. The mean value of all the optimization holding time values in the evaluation period is calculated to obtain an optimization situation value. The number proportion value of the optimization holding time values exceeding the preset optimization holding time threshold in the evaluation period is calculated and is marked as an optimization performance value. In the interval length of the generation of the adjacent recommendation optimization low emergency signal in the evaluation period, the number of the generation of the recommendation optimization high emergency signal is marked as a high emergency duration value, and the maximum high emergency duration value in the evaluation period is marked as a high emergency holding amplitude value. The optimization management evaluation coefficient is calculated by weighted summation of the optimization evaluation abnormality detection value, the optimization situation value, the optimization performance value and the high emergency holding amplitude value. If the optimization management evaluation coefficient exceeds the preset optimization management evaluation coefficient threshold, an optimization evaluation abnormality signal is generated, otherwise, an optimization evaluation qualified signal is generated.

[0014] As a further improvement of the present application: the calculation of the recommendation feedback coefficient includes: Mark all users involved in commodity recommendation during the mark detection period and mark as i, i is a natural number greater than 1; Collect the purchase rate, collection rate and click rate of user i for all recommendation results, and give a preset weight coefficient, and calculate the weighted sum to obtain the recommendation feedback coefficient of user i Compared with the prior art, the beneficial effects of the present application are: The present application can comprehensively and accurately collect and analyze multi-dimensional behavior data of users, deeply mine potential features of commodities, and generate a commodity recommendation list highly consistent with current needs and interest preferences of users in combination with real-time situational information. On this basis, through fine integration and optimization of the recommendation list, the accuracy and pertinence of commodity recommendation can be significantly improved; in addition, the system can also collect feedback of users on the recommendation results in real time to optimize the system in a timely manner. BRIEF DESCRIPTION OF DRAWINGS

[0015] Fig. 1 The system framework diagram of the commodity recommendation system of the electronic mall of the present application.

[0016] Fig. 2 The optimization framework diagram of the commodity recommendation system of the electronic mall of the present application.

[0017] Fig. 3 The flowchart of the commodity recommendation method of the electronic mall of the present application. DETAILED DESCRIPTION

[0018] To make the purpose, technical scheme and advantages of the present application clearer, the technical scheme of the present application will be described clearly and completely below in combination with specific embodiments of the present application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0019] In order to solve the technical problems in the prior art, the present application will be further described in combination with the drawings and embodiments: As Figs. 1 to 3 shown, the present application discloses a commodity recommendation system of an electronic mall, which comprises a user multi-dimensional behavior insight module, a commodity feature deep analysis module, a situational awareness recommendation engine module, a recommendation result integration output module, a dynamic tracking feedback module and an electronic mall supervision end.

[0020] The running process of the user multi-dimensional behavior insight module is as follows: Firstly, by embedding data collection codes in various pages and functional modules of the e-commerce platform, real-time user behavior data is collected, including browsing records, search records, collection records, purchase records, dwell time, and click behavior, etc.

[0021] Then, data mining and machine learning algorithms are used to analyze the user behavior data in depth, and the user's potential interest points and purchase preferences are mined, for example, by analyzing the user's browsing records and dwell time, the user's attention to different categories of goods is determined; by analyzing the user's search keywords, the user's immediate needs are understood; finally, the user behavior analysis results are sent to the context-aware recommendation engine module in the form of user behavior feature vectors.

[0022] The user multi-dimensional behavior insight module is used to collect and analyze various behavior data of users in the e-commerce platform, including but not limited to browsing records, search records, collection records, purchase records, dwell time, click behavior, etc., to deeply understand the user's interest preferences and purchase intentions from multiple dimensions, and send the user behavior analysis results to the context-aware recommendation engine module, which can comprehensively and accurately capture the user's behavior characteristics, deeply understand the user's interest preferences and purchase intentions, and provide more accurate basis for subsequent product recommendations, improving the accuracy and personalization of recommendations.

[0023] The running process of the product feature deep analysis module is as follows: The relevant information of the product is obtained from the product database, and the natural language processing technology and data mining algorithm are used to analyze the semantic and sentiment of the product description text and user evaluation, and the key features of the product and the user's attention are extracted; at the same time, combined with the sales data and inventory of the product, the market performance and popularity of the product are analyzed, for example, by analyzing the keywords in the user evaluation, the user's evaluation of the advantages and disadvantages of the product is understood; by analyzing the sales data, the best-selling degree and seasonal characteristics of the product are determined; finally, the product feature analysis results are sent to the context-aware recommendation engine module in the form of product feature vectors, and are associated with the user behavior feature vectors.

[0024] The product feature deep analysis module conducts comprehensive and in-depth analysis of the features of the products in the e-commerce platform, including the basic attributes of the products (such as brand, model, specification, price, etc.), functional characteristics, user evaluation, sales data, etc., and mines the potential value and selling points of the products, and sends the product feature analysis results to the context-aware recommendation engine module, which can deeply mine the potential value and selling points of the products, provide more abundant information for product recommendations, make the recommendation results more attractive and persuasive, and improve the user's purchase willingness.

[0025] The specific operation process of the context-aware recommendation engine module is as follows: First, the user behavior feature vector and the commodity feature vector are obtained from the user multi-dimensional behavior insight module and the commodity feature depth analysis module, and then the current context information is collected, such as the user's location information obtained through the positioning function of the device, the local weather condition obtained through the weather interface, etc. Then, according to these information, the combination of various recommendation algorithms such as collaborative filtering algorithm, content-based recommendation algorithm and deep learning recommendation algorithm is used to calculate the interest degree of the user to different commodities and generate a personalized commodity recommendation list, for example, during the holiday, combined with the user's purchase history and current location, the commodities suitable for the holiday atmosphere in the region are recommended; in rainy days, related commodities such as rain gear are recommended.

[0026] The context-aware recommendation engine module generates a personalized commodity recommendation list for the user according to the user behavior analysis result and the commodity feature analysis result, combined with the current context information (such as time, place, weather, user device, etc.), and uses the corresponding recommendation algorithm to generate a personalized commodity recommendation list for the user and transmits it to the recommendation result integration output module, which can fully consider the user's current situation and personalized needs, generate a commodity recommendation list that better meets the user's actual needs, improve the real-time and effectiveness of the recommendation, and increase the user's purchase conversion rate.

[0027] The recommendation result integration output module integrates and optimizes the received commodity recommendation list and outputs the integrated and optimized recommendation result to the corresponding page of the electronic mall for the user to browse and select, which can effectively integrate and optimize the recommendation result to display it to the user in a suitable way and provide related recommendation explanation and guidance information, improve the user's acceptance of the recommendation result and purchase willingness, and improve the user's shopping experience.

[0028] Specifically, the recommendation result integration output module obtains a personalized commodity recommendation list from the context-aware recommendation engine module, integrates and sorts the recommendation list, removes duplicate commodities, and optimizes it according to the recommendation score of the commodity and the user's historical preference.

[0029] Then design the recommendation display interface to present the recommended commodities to the user in an intuitive way and provide detailed information, price and evaluation of the commodities and other related information, and provide recommendation explanation and guidance information for each recommended commodity, such as "according to your recent browsing records, we recommend this commodity for you", etc. to help the user understand the reason for the recommendation, and finally output the integrated and optimized recommendation result to the corresponding page of the electronic mall for the user to browse and select.

[0030] The dynamic tracking feedback module tracks the recommendation results, collects feedback information of the user on the recommendation results, and sends all collected feedback information of the recommendation results to the electronic market supervision end, so as to track and collect feedback information of the user, which is beneficial to dynamically optimizing and adjusting the recommendation algorithm and model in time according to the feedback information, so that the recommendation system can continuously adapt to changes in user demand and market environment, and improve the accuracy of recommendation and user satisfaction.

[0031] In some embodiments, as shown in Fig. 2 The dynamic tracking feedback module is communicatively connected to the recommendation optimization urgency decision module, and the specific operation process of the recommendation optimization urgency decision module is as follows: All users involved in the commodity recommendation of the electronic market in the detection period are obtained, and the corresponding user is marked as i, and i is a natural number greater than 1; the feedback information of all recommendation results involved in the user i in the detection period is called from the dynamic tracking feedback module, and the purchase rate, the collection rate and the click rate of the user i for all recommendation results in the detection period are collected.

[0032] The purchase rate, the collection rate and the click rate are weighted and summed to obtain the recommendation feedback coefficient of the user i, that is, the purchase rate, the collection rate and the click rate are respectively assigned to the corresponding preset weight coefficient, and the purchase rate, the collection rate and the click rate are respectively multiplied by the corresponding preset weight coefficient, and the sum of the three groups of product results is marked as the recommendation feedback coefficient; it should be noted that the smaller the value of the recommendation feedback coefficient, the worse the overall satisfaction of the user i for the recommendation results in the detection period.

[0033] The recommendation feedback coefficient is compared with the preset recommendation feedback coefficient threshold value, if the recommendation feedback coefficient exceeds the preset recommendation feedback coefficient threshold value, it indicates that the overall satisfaction of the user i for the recommendation results in the detection period is good, and the user i is marked as a good feedback object; otherwise, it indicates that the overall satisfaction of the user i for the recommendation results in the detection period is not good, and the user i is marked as a poor feedback object.

[0034] The number of poor feedback objects in the detection period is obtained, and the ratio of the number of poor feedback objects to the total number of users involved in the detection period is calculated to obtain a poor feedback summary value, and the poor feedback summary value is compared with the preset poor feedback summary threshold value, if the poor feedback summary value exceeds the preset poor feedback summary threshold value, it indicates that the recommendation system needs to be optimized in time to improve the recommendation accuracy, and a recommendation optimization high urgency signal is generated.

[0035] If the poor feedback summary value does not exceed the preset poor feedback summary threshold value, the recommended feedback coefficients of all users involved in the detection period are averaged to obtain a recommended feedback table value, and the recommended feedback table value is compared with a preset recommended feedback table threshold value. If the recommended feedback table value does not exceed the preset recommended feedback table threshold value, it indicates that the recommendation system needs to be optimized in a timely manner to improve its recommendation accuracy, and a recommended optimization high emergency signal is generated. If the recommended feedback table value exceeds the preset recommended feedback table threshold value, it indicates that the recommendation performance of the commodity recommendation system in the detection period is good, and optimization is not needed in a timely manner, and a recommended optimization low emergency signal is generated.

[0036] The dynamic tracking feedback module sends the feedback information for all the recommended results to the recommendation optimization emergency decision module. The recommendation optimization emergency decision module is used to set a detection period, and preferably the detection period is seven days. The recommended feedback performance in the detection period is analyzed, and a recommended optimization high emergency signal or a recommended optimization low emergency signal is generated through the analysis.

[0037] The recommended optimization high emergency signal or the recommended optimization low emergency signal is sent to the electronic mall supervision end. When the electronic mall supervision end receives the recommended optimization high emergency signal, it issues a corresponding early warning to remind the supervisor to optimize the commodity recommendation algorithm in a timely manner, so that the system can continuously adapt to market changes and user demand evolution, and continuously improve the recommendation performance and effect.

[0038] In some embodiments, as shown in FIG. 13, Fig. 2 The specific analysis process of the optimization management evaluation module is as follows: The number of times that the recommended optimization high emergency signal is generated in the evaluation period is obtained and marked as an optimization evaluation abnormal detection value. The optimization evaluation abnormal detection value is compared with a preset optimization evaluation abnormal detection threshold value. If the optimization evaluation abnormal detection value exceeds the preset optimization evaluation abnormal detection threshold value, it indicates that the optimization management performance for the commodity recommendation system in the detection period is poor, and an optimization evaluation abnormal signal is generated.

[0039] If the optimization evaluation abnormal detection value does not exceed the preset optimization evaluation abnormal detection threshold value, the moment when the recommended optimization high emergency signal is generated is marked as an abnormal generation moment, and the moment when the management personnel takes optimization improvement measures is marked as an optimization moment. The time difference between the optimization moment and the abnormal generation moment is calculated to obtain an optimization holding time value. The greater the optimization holding time value, the less timely the optimization for the corresponding recommended optimization high emergency signal.

[0040] The optimization time value in the evaluation period is obtained and the average value is calculated to obtain the optimization time value, and the number of optimization time values exceeding the preset optimization time threshold in the evaluation period is marked as the optimization performance value, and the number of recommended optimization high-urgency signals involved in the interval time length between the generation time of the adjacent two groups of recommended optimization low-urgency signals in the evaluation period is marked as the high-urgency duration value, and the maximum value of the high-urgency duration value in the evaluation period is marked as the high-urgency amplitude value.

[0041] The optimization management evaluation coefficient is calculated by weighted summation of the optimization evaluation anomaly detection value, the optimization time value, the optimization performance value and the high-urgency amplitude value; that is, the optimization evaluation anomaly detection value, the optimization time value, the optimization performance value and the high-urgency amplitude value are respectively assigned with corresponding preset weight coefficients, and the optimization evaluation anomaly detection value, the optimization time value, the optimization performance value and the high-urgency amplitude value are respectively multiplied by the corresponding preset weight coefficients, and the sum of the four product results is marked as the optimization management evaluation coefficient; it should be noted that the larger the value of the optimization management evaluation coefficient, the worse the comprehensive optimization management performance of the commodity recommendation system in the detection period.

[0042] The optimization management evaluation coefficient is compared with the preset optimization management evaluation coefficient threshold value, if the optimization management evaluation coefficient exceeds the preset optimization management evaluation coefficient threshold value, it indicates that the comprehensive optimization management performance of the commodity recommendation system in the detection period is poor, and an optimization evaluation anomaly signal is generated; if the optimization management evaluation coefficient does not exceed the preset optimization management evaluation coefficient threshold value, it indicates that the comprehensive optimization management performance of the commodity recommendation system in the detection period is good, and an optimization evaluation qualified signal is generated.

[0043] The recommended optimization urgency decision module sends the recommended optimization high-urgency signal or the recommended optimization low-urgency signal to the optimization management evaluation module, and the optimization management evaluation module is used to set an evaluation period, preferably, the evaluation period is one hundred and forty days; the recommended optimization management performance in the evaluation period is analyzed, and the optimization evaluation qualified signal or the optimization evaluation anomaly signal is generated by analysis.

[0044] The optimization evaluation qualified signal or the optimization evaluation anomaly signal is sent to the electronic mall supervision end, and the electronic mall supervision end sends a corresponding early warning when receiving the optimization evaluation anomaly signal, so as to remind the supervisor to timely strengthen the system optimization management, ensure the optimization scientificity and effectiveness in the future, and further maintain the recommendation performance and effect of the recommendation system.

[0045] As shown in Fig. 3 The present application further provides a commodity recommendation method of an electronic mall, which comprises the following steps: Step one, collect and analyze various behavior data of users in the electronic mall, and deeply understand the interest preference and purchase intention of users from multiple dimensions; Step two, comprehensive and in-depth feature analysis of the goods in the electronic mall, mining the potential value and selling points of the goods; Step three, according to the user behavior analysis results and the commodity feature analysis results, combining the current situation information, using the corresponding recommendation algorithm to generate personalized commodity recommendation list for the user; Step four, integrate and optimize the commodity recommendation list, output the integrated and optimized recommendation results to the corresponding page of the electronic mall for the user to browse and select; Step five, track the recommendation results and collect the feedback information of the user on the recommendation results.

[0046] All users involved in the commodity recommendation of the electronic mall in the detection period are obtained, and the corresponding user is marked as i, and i is a natural number greater than 1; the feedback information of all recommendation results involved by user i in the detection period is called from the dynamic tracking feedback module, and the purchase rate, collection rate and click rate of user i for all recommendation results in the detection period are collected.

[0047] The purchase rate, collection rate and click rate are weighted and summed to obtain the recommendation feedback coefficient of user i, that is, the purchase rate, collection rate and click rate are respectively assigned to the corresponding preset weight coefficient, and the purchase rate, collection rate and click rate are respectively multiplied by the corresponding preset weight coefficient, and the sum of the three groups of product results is marked as the recommendation feedback coefficient; It should be noted that the smaller the value of the recommendation feedback coefficient, the worse the overall satisfaction of user i for the recommendation results in the detection period.

[0048] The recommendation feedback coefficient is compared with the preset recommendation feedback coefficient threshold value, if the recommendation feedback coefficient exceeds the preset recommendation feedback coefficient threshold value, it indicates that the overall satisfaction of user i for the recommendation results in the detection period is better, then user i is marked as a good feedback object; Otherwise, it indicates that the overall satisfaction of user i for the recommendation results in the detection period is not good, and user i is marked as a poor feedback object.

[0049] The number of poor feedback objects in the detection period is obtained and compared with the total number of users involved in the detection period to obtain the poor feedback summary value, and the poor feedback summary value is compared with the preset poor feedback summary threshold value, if the poor feedback summary value exceeds the preset poor feedback summary threshold value, it indicates that the recommendation system needs to be optimized in time to improve its recommendation accuracy, then a recommendation optimization high emergency signal is generated.

[0050] If the poor feedback summary value does not exceed the preset poor feedback summary threshold value, the recommended feedback coefficients of all users involved in the detection period are averaged to obtain a recommended feedback table value, and the recommended feedback table value is compared with a preset recommended feedback table threshold value. If the recommended feedback table value does not exceed the preset recommended feedback table threshold value, it indicates that the recommendation system needs to be optimized in a timely manner to improve its recommendation accuracy, and a recommendation optimization high-urgency signal is generated. If the recommended feedback table value exceeds the preset recommended feedback table threshold value, it indicates that the recommendation performance of the commodity recommendation system in the detection period is good, and timely optimization is not required, and a recommendation optimization low-urgency signal is generated.

[0051] In some embodiments, the recommendation optimization urgency decision module is communicatively connected to the optimization management evaluation module. The recommendation optimization urgency decision module sends the recommendation optimization high-urgency signal or the recommendation optimization low-urgency signal to the optimization management evaluation module. The optimization management evaluation module is used to set an evaluation period, which is preferably 140 days. The optimization management performance in the evaluation period is analyzed, and an optimization evaluation qualified signal or an optimization evaluation abnormal signal is generated by the analysis. The optimization evaluation qualified signal or the optimization evaluation abnormal signal is sent to the electronic mall supervision end. When the electronic mall supervision end receives the optimization evaluation abnormal signal, it issues a corresponding early warning to remind the supervisor to timely strengthen system optimization management, ensure the scientific nature and effectiveness of subsequent optimization, and further maintain the recommendation performance and effect of the recommendation system. The specific analysis process of the optimization management evaluation module is as follows: The number of times that the recommendation optimization high-urgency signal is generated in the evaluation period is obtained and marked as an optimization evaluation abnormal detection value. The optimization evaluation abnormal detection value is compared with a preset optimization evaluation abnormal detection threshold value. If the optimization evaluation abnormal detection value exceeds the preset optimization evaluation abnormal detection threshold value, it indicates that the optimization management performance for the commodity recommendation system in the detection period is poor, and an optimization evaluation abnormal signal is generated. If the optimization evaluation abnormal detection value does not exceed the preset optimization evaluation abnormal detection threshold value, the time when the recommendation optimization high-urgency signal is generated is marked as the abnormal generation time, and the time when the management personnel takes optimization improvement measures is marked as the optimization time. The time difference between the optimization time and the abnormal generation time is calculated to obtain an optimization holding time value. The larger the optimization holding time value, the less timely the optimization for the corresponding recommendation optimization high-urgency signal. All optimization holding time values in the evaluation period are obtained and averaged to obtain an optimization time value. The proportion of the number of optimization holding time values that exceed the preset optimization holding time threshold value in the evaluation period is marked as an optimization performance value. The number of generated recommendation optimization high-urgency signals in the interval time length between the generation times of adjacent two groups of recommendation optimization low-urgency signals in the evaluation period is marked as a high-urgency duration value, and the largest high-urgency duration value in the evaluation period is marked as a high-urgency holding amplitude value. The optimization management evaluation coefficient is calculated by weighted summation of the optimization evaluation abnormality value, the optimization situation value, the optimization performance value and the high emergency holding range value; that is, the optimization evaluation abnormality value, the optimization situation value, the optimization performance value and the high emergency holding range value are respectively assigned corresponding preset weight coefficients, and the optimization evaluation abnormality value, the optimization situation value, the optimization performance value and the high emergency holding range value are respectively multiplied by the corresponding preset weight coefficients, and the sum of the four product results is marked as the optimization management evaluation coefficient; it should be noted that the larger the value of the optimization management evaluation coefficient, the worse the comprehensive optimization management performance of the commodity recommendation system in the detection period; The optimization management evaluation coefficient is compared with the preset optimization management evaluation coefficient threshold value, if the optimization management evaluation coefficient exceeds the preset optimization management evaluation coefficient threshold value, it indicates that the comprehensive optimization management performance of the commodity recommendation system in the detection period is poor, and an optimization evaluation abnormality signal is generated; if the optimization management evaluation coefficient does not exceed the preset optimization management evaluation coefficient threshold value, it indicates that the comprehensive optimization management performance of the commodity recommendation system in the detection period is good, and an optimization evaluation qualified signal is generated.

[0052] The main functions of the application are as follows: 1、In the application, by combining and associating multiple modules, multi-dimensional behavior data of users can be comprehensively and accurately collected and analyzed, potential features of commodities can be deeply mined, commodity recommendation lists that meet current needs and interests of users can be generated in combination with real-time situational information, and the recommendation lists can be carefully integrated and optimized, so that the accuracy and pertinence of commodity recommendation are effectively improved, real-time feedback of users on the recommendation results can be collected to optimize the system in time, and the recommendation performance and effect are continuously improved, which shows significant advantages in meeting personalized needs of users, improving sales efficiency of the mall and improving user shopping experience, etc. 2、In the application, the recommendation feedback performance in the detection period is analyzed by the recommendation optimization emergency decision module, the commodity recommendation algorithm, etc. are optimized when the recommendation optimization high emergency signal is generated, so that the system can continuously adapt to market changes and evolution of user needs, and the recommendation performance and effect are continuously improved, and the recommendation optimization management performance in the evaluation period is analyzed by the optimization management evaluation module, the system optimization management is strengthened when the optimization evaluation abnormality signal is generated, the subsequent optimization is ensured to be scientific and effective, the recommendation performance and effect of the recommendation system are further maintained, and the intelligent level is high.

[0053] In summary, after reading the application document, those skilled in the art can make various corresponding transformation schemes according to the technical solutions and technical concepts of the application without creative mental labor, which all belong to the scope protected by the application.

Claims

1. A product recommendation system for an e-commerce platform, characterized in that, include: The user multidimensional behavior insight module is used to collect and analyze user behavior data and output user behavior feature vectors. The product feature deep analysis module is used to acquire and analyze product information and output product feature vectors; The context-aware recommendation engine module is used to obtain user behavior feature vectors and product feature vectors, collect current context information, and generate a product recommendation list. The recommendation result integration and output module is used to receive the product recommendation list, integrate and optimize it, and generate a recommendation display interface; The recommendation optimization urgency decision module is used to obtain and analyze feedback information from all recommendation results and generate recommendation optimization signals.

2. The product recommendation system for an e-commerce platform according to claim 1, characterized in that, It also includes a dynamic tracking feedback module, which collects user feedback on the recommendation results, and a communication connection to the recommendation optimization urgency decision-making module.

3. The product recommendation system for an e-commerce platform according to claim 1, characterized in that, The step of generating the recommendation optimization signal by the recommendation optimization urgency decision module includes: Obtain all users involved in product recommendations during the detection period, labeled as i, where i is a natural number greater than 1; During the detection period, the purchase rate, collection rate, and click rate of user i for all recommended results were collected, and the corresponding weight coefficients were determined. The weighted sum was then used to calculate the recommendation feedback coefficient. If the recommendation feedback coefficient exceeds the preset recommendation feedback coefficient threshold, user i will be marked as a good feedback object; otherwise, user i will be marked as a bad feedback object. Calculate the summary value of negative feedback, which is the ratio of the number of negative feedback objects to the total number of users involved in the detection period. If the total negative feedback value exceeds the preset negative feedback threshold, a recommended optimization high-urgent signal will be generated.

4. The product recommendation system for an e-commerce platform according to claim 3, characterized in that, The step of generating the recommendation optimization signal by the recommendation optimization urgency decision module further includes: If the total negative feedback value does not exceed the preset negative feedback threshold, calculate the average of the recommendation feedback coefficients of all users to obtain the recommendation feedback table value; If the recommended feed value does not exceed the preset recommended feed threshold, a recommended optimized high-urgency signal is generated. If the recommended feed value exceeds the preset recommended feed threshold, a recommended optimized low emergency signal is generated.

5. A product recommendation system for an e-commerce platform according to claim 3 or 4, characterized in that, The recommended optimization of the urgency decision-making module's communication connection to the management and evaluation module is optimized. The optimization management evaluation module is used to obtain the number of times the high emergency signal is recommended for optimization during the evaluation period and mark it as the optimization evaluation abnormal detection value; If the abnormal detection value of the optimization evaluation exceeds the preset abnormal detection threshold of the optimization evaluation, an abnormal signal of the optimization evaluation is generated; If the optimized assessment anomaly detection value does not exceed the preset optimized assessment anomaly detection threshold, the time when the recommended optimized high emergency signal is generated will be marked as the anomaly time, and the time when optimization and improvement measures are taken will be marked as the optimization time.

6. The product recommendation system for an e-commerce platform according to claim 5, characterized in that, The optimization management evaluation module is also used for The optimization duration value is obtained by calculating the time difference between the optimization time and the time of occurrence. The average of all optimization duration values ​​within the evaluation period is then calculated to obtain the optimization time value. The percentage of optimized duration values ​​exceeding the preset optimized duration threshold during the evaluation period is marked as the optimal performance value. During the assessment period, within the interval between adjacent recommended optimization of low emergency signal generation, the number of recommended optimization of high emergency signals generated is marked as the high emergency duration value, and the high emergency duration value with the largest value during the assessment period is marked as the high emergency amplitude value. The optimized management evaluation coefficient is obtained by weighting and summing the optimized evaluation abnormal detection value, optimized situation value, excellent performance value and high emergency amplitude value.

7. The product recommendation system for an e-commerce platform according to claim 6, characterized in that, The optimization management evaluation module is also used for If the optimization management evaluation coefficient exceeds the preset optimization management evaluation coefficient threshold, an optimization evaluation abnormality signal will be generated. If the optimization management evaluation coefficient does not exceed the preset optimization management evaluation coefficient threshold, an optimization evaluation qualified signal will be generated.

8. A product recommendation method for an e-commerce platform, characterized in that, include: Analyze user behavior data to obtain user behavior feature vectors, analyze product information to obtain product feature vectors, and combine current context information to generate a product recommendation list; Collect user feedback on the recommendation results and calculate the recommendation feedback coefficient to determine the best and worst feedback objects; The percentage of poorly fed objects is calculated to obtain the total poorly fed value. If the total poorly fed value exceeds the preset poorly fed value threshold, a high-urgency signal for recommendation optimization is generated. Otherwise, the mean value of the recommendation feedback coefficient is calculated to obtain the recommendation feedback table value. When the recommended feed value exceeds the preset recommended feed threshold, a recommended optimized low-urgency signal is generated; otherwise, a recommended optimized high-urgency signal is generated.

9. The product recommendation method for an e-commerce platform according to claim 8, characterized in that, Also includes: The number of times a recommended optimized high-urgency signal was generated during the evaluation period was obtained and marked as an optimized evaluation anomaly detection value; If the optimized evaluation abnormal detection value exceeds the preset optimized evaluation abnormal detection threshold, an optimized evaluation abnormal signal is generated; otherwise, the time when the recommended optimized high emergency signal is generated is marked as the abnormal time, and the time when optimization and improvement measures are taken is marked as the optimization time. The time difference between the optimization time and the abnormal time is calculated to obtain the optimization duration value. The optimal time-state value is obtained by calculating the mean of all optimal time-state values ​​during the evaluation period; Calculate the percentage of optimized duration values ​​that exceed the preset optimized duration threshold during the evaluation period and mark them as optimal performance values; During the assessment period, within the interval between adjacent recommended optimization of low emergency signal generation, the number of recommended optimization of high emergency signals generated is marked as the high emergency duration value, and the high emergency duration value with the largest value during the assessment period is marked as the high emergency amplitude value. The optimized management evaluation coefficient is obtained by weighted summation of the optimized evaluation abnormal detection value, optimized time situation value, excellent performance value and high emergency holding amplitude value. If the optimization management evaluation coefficient exceeds the preset optimization management evaluation coefficient threshold, an optimization evaluation abnormality signal will be generated; otherwise, an optimization evaluation qualified signal will be generated.

10. A product recommendation method for an e-commerce platform according to claim 8 or 9, characterized in that, The calculation of the recommendation feedback coefficient includes: All users involved in product recommendations during the detection period are labeled as i, where i is a natural number greater than 1; The purchase rate, collection rate, and click rate of user i for all recommended results are collected, and a preset weight coefficient is assigned. The weighted sum is then calculated to obtain the recommendation feedback coefficient of user i.

Images