E-commerce platform commodity recommendation method and system based on big data

Abstract

The invention relates to the technical field of commodity recommendation, in particular to an e-commerce platform commodity recommendation method and system based on big data. The method comprises the following steps: firstly, collecting and storing user behavior data of a user on an e-commerce platform; secondly, constructing a mistaken touch feature extraction model, and extracting various features related to mistaken touch according to the user behavior data to obtain mistaken touch features; then, on the basis of the mistaken touch features, mistaken touch judgment conditions are set, mistaken touch behavior labeling is carried out, and mistaken touch behavior labeling data are obtained; then, constructing and training an e-commerce platform mistaken touch detection model, and performing mistaken touch detection on the click behavior of the user according to the mistaken touch features and the mistaken touch behavior annotation data; and finally, processing the mistaken touch behavior of the user according to the mistaken touch detection result of the mistaken touch detection model of the e-commerce platform so as to adjust a commodity recommendation strategy of the e-commerce platform. According to the invention, accurate false touch detection of the click behavior of the user is realized through the false touch detection model of the e-commerce platform.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity recommendation, and in particular to a commodity recommendation method and system for an e-commerce platform based on big data. Background Art

[0002] Due to the rapid development of modern technology, major e-commerce platforms have emerged one after another. The product recommendation system in the e-commerce platform intelligently provides users with personalized product recommendations by analyzing the user's historical behavior, interest preferences and purchasing habits. This not only improves the user experience and helps them find products of interest more quickly, but also greatly improves the platform's conversion rate and sales. Through accurate recommendations, the platform can improve user stickiness and satisfaction, while enhancing product exposure, optimizing inventory management, and promoting merchants to achieve precision marketing, ultimately promoting the healthy development of the entire e-commerce ecosystem.

[0003] However, the problem of accidental touches also arises, especially on touch screen devices, where users may accidentally touch the interface and click on products, resulting in recommendations that are irrelevant to their actual needs and / or interests, thus affecting the user experience and the accuracy of product recommendations. Therefore, accurate accidental touch detection can help accurately identify and eliminate accidental touch data, improve the accuracy of product recommendations, avoid affecting the user's shopping experience, and help the platform optimize resource allocation and improve user satisfaction.

[0004] To this end, a product recommendation method and system for an e-commerce platform based on big data is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an e-commerce platform product recommendation method and system based on big data. First, the present invention collects and stores the user behavior data of users on the e-commerce platform; secondly, constructs a false touch feature extraction model, extracts multiple features related to false touches according to the user behavior data, and obtains false touch features; then, based on the false touch features, sets false touch judgment conditions and performs false touch behavior annotation to obtain false touch behavior annotation data; then, constructs and trains an e-commerce platform false touch detection model, performs false touch detection on the user's click behavior according to the false touch features and the false touch behavior annotation data; finally, processes the user's false touch behavior according to the false touch detection results of the e-commerce platform false touch detection model to adjust the e-commerce platform product recommendation strategy. The present invention realizes accurate false touch detection of user click behavior through the e-commerce platform false touch detection model.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A product recommendation method for an e-commerce platform based on big data, comprising:

[0008] Collect and store user behavior data on e-commerce platforms, including: click behavior data, page interaction data, user history data, device and network data, and contextual data;

[0009] Constructing an accidental touch feature extraction model, extracting multiple features related to the accidental touch according to the user behavior data to obtain the accidental touch features; the accidental touch features include: dwell time features, click frequency features, click position features, click and product matching features, device type features and network type features;

[0010] Based on the accidental touch characteristics, the accidental touch judgment conditions are set and the accidental touch behavior is marked to obtain the accidental touch behavior marking data; the accidental touch judgment conditions include: short-time stay judgment, frequent click judgment, abnormal click position judgment and click and historical behavior inconsistency judgment;

[0011] Constructing an e-commerce platform accidental touch detection model, performing accidental touch detection on the user's click behavior according to the accidental touch characteristics and the accidental touch behavior annotation data, and obtaining an accidental touch probability;

[0012] Training and optimizing the parameters of the e-commerce platform false touch detection model, and evaluating the false touch detection performance of the e-commerce platform false touch detection model for user click behaviors;

[0013] The user's accidental touch behavior is processed according to the accidental touch detection result of the e-commerce platform accidental touch detection model to adjust the product recommendation strategy of the e-commerce platform.

[0014] Preferably, the click behavior data includes: click time, click product ID, click location coordinates and click page area; the page interaction data includes: page dwell time, scrolling behavior and touch events; the user history data includes: browsing history, shopping cart, search history and purchase history; the device and network data includes: device type, operating system, browser type, screen resolution and network type; the context data includes: a product list generated by a recommendation algorithm.

[0015] Preferably, the dwell time feature includes: calculating the dwell time of the user on the product page after clicking on the product; the click frequency feature includes: recording the number of times the user clicks on the same and / or adjacent products within a predetermined time; the click position feature includes: recording the page position coordinates of the user's click and the distance from the product display area; the click and product matching feature includes: comparing the user's browsing history with the matching degree of the currently clicked product, and calculating the mismatched click behavior; the device type feature includes: obtaining the type of device used by the user for clicking, including PC, mobile phone and tablet, and judging the impact of different device types on accidental touches; the network type feature includes: obtaining the network used by the user for clicking, and judging the impact of different network types on accidental touches.

[0016] Preferably, the method for determining the short-time stay is: setting a page stay time threshold, if the user stays on the product page for less than the page stay time threshold after clicking on the product, it is marked as an accidental touch; if the user stays on the product page for greater than and / or equal to the page stay time threshold after clicking on the product, it is marked as a normal click;

[0017] The method for determining frequent clicks is: setting a click count threshold, if the number of times a user clicks on the same and / or adjacent products within a predetermined time is greater than the click count threshold, it is marked as an accidental touch; if the number of times a user clicks on the same and / or adjacent products within a predetermined time is less than and / or equal to the click count threshold, it is marked as a normal click;

[0018] The method for determining abnormal click position is as follows: setting a page edge area threshold and a distance threshold, if the position coordinates clicked by the user are within the page edge area threshold and the distance from the product display area is greater than the distance threshold, it is marked as an accidental touch; if the position coordinates clicked by the user are outside the page edge area threshold and the distance from the product display area is less than and / or equal to the distance threshold, it is marked as a normal click;

[0019] The method for determining whether a click is inconsistent with historical behavior includes: matching the category of the product clicked by the user with the category of the products he has browsed in the past; if the type of the product clicked by the user does not match the category of the products he has browsed in the past, marking it as an accidental touch; if the type of the product clicked by the user matches the category of the products he has browsed in the past, marking it as a normal click.

[0020] Preferably, the e-commerce platform mistouch detection model is constructed by an LSTM model, and the LSTM model includes: an input layer, an LSTM layer, a mistouch detection layer, and a mistouch probability output layer; wherein the input layer includes: combining the user behavior data, the time step, and the feature dimension of the mistouch feature to form an input time series matrix, which is input into the LSTM layer;

[0021] The LSTM layer includes: extracting the behavior pattern in the time series through the LSTM model, processing the long-term dependency information required for false touch detection, and obtaining the hidden state; the false touch detection layer includes: mapping the hidden state obtained by the LSTM layer to the probability value of false touch through the fully connected layer; the false touch probability output layer includes: outputting the false touch probability through the activation function.

[0022] Preferably, the training and optimization of the e-commerce platform false touch detection model includes: dividing the user behavior data into a training set and a validation set; using the training set to optimize the e-commerce platform false touch detection model by calculating the false touch detection loss function and adjusting the batch size, time step and optimizer parameters of the e-commerce platform false touch detection model; using the optimized e-commerce platform false touch detection model to perform false touch detection on user behavior data; when a false touch is detected, automatically removing the product from the recommendation list to avoid affecting subsequent recommendations; using the validation set to evaluate the user click false touch detection performance of the e-commerce platform false touch detection model.

[0023] A product recommendation system for an e-commerce platform based on big data, comprising:

[0024] User behavior data acquisition module, used to collect and store user behavior data on the e-commerce platform, including: click behavior data, page interaction data, user history data, device and network data, and context data;

[0025] An accidental touch feature extraction module is used to construct an accidental touch feature extraction model, extract multiple features related to the accidental touch according to the user behavior data, and obtain the accidental touch features; the accidental touch features include: dwell time features, click frequency features, click position features, click and product matching features, device type features, and network type features;

[0026] An error touch condition setting module is used to set the error touch judgment condition and mark the error touch behavior based on the error touch feature to obtain the error touch behavior marking data; the error touch judgment condition includes: short-time stay judgment, frequent click judgment, abnormal click position judgment and click and historical behavior inconsistency judgment;

[0027] An accidental touch behavior detection module is used to build an accidental touch detection model for an e-commerce platform, and to perform accidental touch detection on the user's click behavior according to the accidental touch characteristics and the accidental touch behavior annotation data to obtain an accidental touch probability;

[0028] An error detection model evaluation module is used to train and optimize the parameters of the e-commerce platform error detection model, and evaluate the error detection performance of the e-commerce platform error detection model for user click behaviors;

[0029] The accidental touch behavior processing module processes the user's accidental touch behavior according to the accidental touch detection result of the accidental touch detection model of the e-commerce platform to adjust the product recommendation strategy of the e-commerce platform.

[0030] Preferably, the dwell time feature includes: calculating the dwell time of the user on the product page after clicking on the product; the click frequency feature includes: recording the number of times the user clicks on the same and / or adjacent products within a predetermined time; the click position feature includes: recording the page position coordinates of the user's click and the distance from the product display area; the click and product matching feature includes: comparing the user's browsing history with the matching degree of the currently clicked product, and calculating the mismatched click behavior; the device type feature includes: obtaining the type of device used by the user for clicking, including PC, mobile phone and tablet, and judging the impact of different device types on accidental touches; the network type feature includes: obtaining the network used by the user for clicking, and judging the impact of different network types on accidental touches.

[0031] Preferably, the method for determining the short-time stay is: setting a page stay time threshold, if the user stays on the product page for less than the page stay time threshold after clicking on the product, it is marked as an accidental touch; if the user stays on the product page for greater than and / or equal to the page stay time threshold after clicking on the product, it is marked as a normal click;

[0032] The method for determining frequent clicks is: setting a click count threshold, if the number of times a user clicks on the same and / or adjacent products within a predetermined time is greater than the click count threshold, it is marked as an accidental touch; if the number of times a user clicks on the same and / or adjacent products within a predetermined time is less than and / or equal to the click count threshold, it is marked as a normal click;

[0033] The method for determining abnormal click position is as follows: setting a page edge area threshold and a distance threshold, if the position coordinates clicked by the user are within the page edge area threshold and the distance from the product display area is greater than the distance threshold, it is marked as an accidental touch; if the position coordinates clicked by the user are outside the page edge area threshold and the distance from the product display area is less than and / or equal to the distance threshold, it is marked as a normal click;

[0034] The method for determining whether a click is inconsistent with historical behavior includes: matching the category of the product clicked by the user with the category of the products he has browsed in the past; if the type of the product clicked by the user does not match the category of the products he has browsed in the past, marking it as an accidental touch; if the type of the product clicked by the user matches the category of the products he has browsed in the past, marking it as a normal click.

[0035] Preferably, the e-commerce platform false touch detection model is constructed by an LSTM model, and the LSTM model includes: an input layer, an LSTM layer, a false touch detection layer and a false touch probability output layer.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. The present invention constructs a false touch feature extraction model by collecting user behavior data, including click behavior data, page interaction data, user history data, device and network data, and context data, and extracting false touch features from the user behavior data; by capturing dwell time features, click frequency features, click location features, click and product matching features, device type features, and network type features, a single click behavior can be placed in a continuous behavior for analysis, thereby accurately characterizing the key influencing factors of the false touch behavior, and providing a multi-dimensional data foundation for the subsequent setting of false touch judgment conditions and the construction of a false touch detection model for the e-commerce platform.

[0038] 2. The present invention performs conditional labeling on user behavior data by setting false touch judgment conditions, marking user behaviors as "normal clicks" and "false touches", wherein the false touch judgment conditions include: short-time stay judgment, frequent click judgment, abnormal click position judgment, and click and historical behavior inconsistency judgment; this labeling method can provide clear label data for the false touch detection model of the e-commerce platform, helping it to accurately identify which clicks are caused by false touch behaviors such as sliding, misoperation, and unintentional clicks, rather than the user's real interests. This labeling method provides accurate false touch judgment information for the training of the false touch detection model of the e-commerce platform, enabling the model to learn the specific characteristics of false touch behaviors from historical data, thereby more efficiently identifying false touch clicks during real-time detection, and improving the false touch detection model of the e-commerce platform. The ability to detect false touch behaviors improves.

[0039] 3. The present invention proposes an e-commerce platform false touch detection model, which analyzes user behavior data through the LSTM model, can capture the erroneous operations behind the false touch behaviors and the click behaviors that do not conform to the normal interaction logic, and effectively identify the false touch patterns in the user behaviors; this e-commerce platform false touch detection model based on the LSTM model can automatically identify the false touch behaviors in the user's click behaviors through deep learning, avoid the limitations of traditional rule detection, and can process complex and long-span user behavior data, thereby improving the false touch detection capability of the e-commerce platform false touch detection model for the user's click behaviors, and improving the false touch detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of a method for recommending products on an e-commerce platform based on big data provided by an embodiment of the present invention;

[0041] Figure 2 A schematic diagram of the structure of a product recommendation system for an e-commerce platform based on big data provided by an embodiment of the present invention;

[0042] Figure 3 A schematic diagram of a flow chart of a short-time stay determination method provided by an embodiment of the present invention;

[0043] Figure 4 A schematic diagram of a process flow of a frequent click determination method provided by an embodiment of the present invention;

[0044] Figure 5 A schematic diagram of a flow chart of a method for determining abnormal click position provided by an embodiment of the present invention;

[0045] Figure 6 A schematic flow chart of a method for determining whether a click is inconsistent with a historical behavior provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] 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.

[0047] Due to the rapid development of modern technology, major e-commerce platforms have emerged one after another. The product recommendation system in the e-commerce platform intelligently provides users with personalized product recommendations by analyzing the user's historical behavior, interest preferences and purchasing habits. However, the problem of accidental touches also comes with it, especially on touch screen devices, where users may accidentally touch the interface and click on the product, resulting in recommendations that are irrelevant to their actual needs or interests, thus affecting the user experience and the accuracy of product recommendations. Therefore, accurate accidental touch detection can help accurately identify and eliminate accidental touch data, improve the accuracy of product recommendations, avoid affecting the user's shopping experience, and help the platform optimize resource allocation and improve user satisfaction.

[0048] The present invention proposes a product recommendation method for an e-commerce platform based on big data, which realizes accurate false touch detection of user click behavior. This method is applied to a product recommendation system for an e-commerce platform based on big data. For the specific method and system flowchart, please refer to Figure 1 and Figure 2 In order to illustrate that the method and system of the present invention can play a role in detecting the user's accidental touch behavior, the effectiveness of the present invention will be described from two embodiments below.

[0049] Embodiment 1

[0050] In the specific implementation process of this embodiment, the method of the present invention is implemented by a specific system module. Figure 1The specific steps of the method proposed in the present invention are described in detail; the specific steps include: collecting and storing user behavior data of users on the e-commerce platform; constructing a false touch feature extraction model, extracting multiple features related to false touches according to the user behavior data, and obtaining false touch features; based on the false touch features, setting false touch judgment conditions and marking false touch behaviors; constructing an e-commerce platform false touch detection model to perform false touch detection on the user's click behavior to obtain a false touch probability; setting a false touch judgment threshold, and performing false touch judgment on the false touch probability according to the false touch judgment threshold; training and optimizing the parameters of the e-commerce platform false touch detection model, and evaluating the false touch detection performance of the e-commerce platform false touch detection model for user click behaviors; processing the user's false touch behavior according to the false touch detection results of the e-commerce platform false touch detection model, so as to adjust the e-commerce platform product recommendation strategy.

[0051] The user behavior data of the present invention is Figure 2 The user behavior data acquisition module in the system shown collects and uses Apache Kafka for real-time data transmission, and transmits the user behavior data to Hadoop HDFS for storage, which is convenient for subsequent analysis and training;

[0052] Furthermore, the user behavior data includes: click behavior data, page interaction data, user history data, device and network data and context data; wherein, the click behavior data includes: click time, click product ID, click location coordinates and click page area; the page interaction data includes: page dwell time, scrolling behavior (scroll up and scroll down) and touch events (strength and direction of scrolling gesture); the user history data includes: browsing history, shopping cart, search history and purchase history; the device and network data includes: device type (PC, mobile phone and tablet), operating system (MacOS, Windows and Linuks), browser type (Chrome browser, Microsoft Edge browser and QQ browser), screen resolution (1920×1080 and 1280×720) and network type (4G, 5G and WiFi); the context data includes: the product list generated by the recommendation algorithm.

[0053] In the embodiment of the present application, the click behavior data provides an accurate description of the click behavior, which helps the system analyze the rationality of the click and determine whether it conforms to the normal user intention; the page interaction data captures the dynamic operation between the user and the page, which helps to identify the false touch behavior caused by hand slipping, sliding too fast or other operations; the user's historical data is an important reference for behavioral preferences, which helps the system to determine the consistency of the current click behavior with the user's real interests and improve the accuracy of false touch detection; the device and network data provide environmental information of the user's operation, which helps to identify the false touch behavior caused by the device or network conditions; the context data associates the current click behavior with the recommendation result, which helps the system to determine whether the click conforms to the logical output of the recommendation algorithm, and further improves the accuracy of false touch detection. By integrating various aspects of user behavior data, the user's click behavior is analyzed in multiple dimensions to identify the possibility of false touch. This data fusion method is more accurate than a single data source, effectively solves the shortcomings of the traditional false touch detection method, and provides a reliable data foundation for the subsequent construction of the false touch feature extraction module.

[0054] Further, by Figure 2 The accidental touch feature extraction module in the system constructs an accidental touch feature extraction model, extracts multiple features related to the accidental touch according to the user behavior data, and obtains the accidental touch features; the accidental touch features include: dwell time features, click frequency features, click position features, click and product matching features, device type features, and network type features;

[0055] The dwell time feature includes: calculating the dwell time of a user on a product page after clicking on the product; the calculation formula for the dwell time is:

[0056] T stay =T leave -T click ;

[0057] Among them, T stay T is the page dwell time; leave The timestamp when the user leaves the page; T click The timestamp of when the user clicked the page;

[0058] The click frequency feature includes: recording the number of times a user clicks on the same and / or adjacent products within a predetermined time; the specific extraction method is: setting the predetermined time to 2 seconds, and counting the number of times a user clicks on the same and / or adjacent products within the predetermined time;

[0059] The click position feature includes: recording the page position coordinates clicked by the user and the distance between the page position coordinates and the product display area; the distance calculation formula is:

[0060]

[0061] Where d is the distance between the page location coordinates where the user clicks and the product display area; x click The horizontal coordinate of the page clicked by the user; y click The vertical coordinate of the page clicked by the user; x center is the horizontal coordinate of the center of the product display area; center The vertical coordinate of the center of the product display area;

[0062] The click-to-product matching feature includes: comparing the matching degree between the user's browsing history and the currently clicked product, and calculating the mismatching click behavior; the matching degree calculation formula is:

[0063]

[0064] Among them, M is the matching degree; H is the product attribute of the user's browsing history; G is the attribute of the currently clicked product;

[0065] The device type feature includes: obtaining the type of device used by the user to click, including PC, mobile phone and tablet, and determining the impact of different device types on accidental touches;

[0066] The network type feature includes: obtaining the network clicked by the user, including Wi-Fi, 4G and 5G, and determining the impact of different network types on false touches.

[0067] In an embodiment of the present application, by constructing a false touch feature extraction model, multi-dimensional false touch features are extracted from user behavior data in order to comprehensively describe the characteristics of false touch behavior. False touches are identified from the user operation level through dwell time features, click frequency features, and click location features; false touches are judged from the environmental level through device type and network type features; and judgment accuracy is improved through click and product matching features combined with user historical interests. By extracting multi-dimensional features, the limitations of single-feature false touch detection are avoided, and different user behavior patterns and scenario requirements are adapted. Through this false touch feature extraction model, the system of the present invention can accurately capture possible false touch patterns and provide high-quality input for subsequent false touch detection and product recommendation system optimization.

[0068] Further, by Figure 2 The mis-touch condition setting module in the system shown sets mis-touch judgment conditions and performs mis-touch behavior annotation based on the mis-touch characteristics to obtain mis-touch behavior annotation data; the mis-touch judgment conditions include: short-time stay judgment, frequent click judgment, abnormal click position judgment, and click and historical behavior inconsistency judgment;

[0069] Among them, see Figure 3The flowchart of the short-time stay determination method shown in the figure is as follows: a page stay time threshold is set, and if the user stays on the product page after clicking on the product is less than the page stay time threshold, it is marked as an accidental touch; if the user stays on the product page after clicking on the product is greater than and / or equal to the page stay time threshold, it is marked as a normal click;

[0070] Specifically, set the page dwell time threshold T threshold is 2 seconds, comparing the page dwell time T after the user clicks on the product stay :

[0071] If T stay <T threshold , marked as accidental touch;

[0072] If T stay ≥T threshold , marked as normal click;

[0073] See also Figure 4 The flowchart of the frequent click determination method shown is as follows: a click count threshold is set, and if the number of times a user clicks on the same and / or adjacent products within a predetermined time is greater than the click count threshold, it is marked as an accidental touch; if the number of times a user clicks on the same and / or adjacent products within a predetermined time is less than and / or equal to the click count threshold, it is marked as a normal click;

[0074] Specifically, set the click count threshold C threshold 5 times, the predetermined time W is 1 minute; count the number of times C that users click on the same and / or adjacent products within the predetermined time W freq :

[0075] If C freq >C threshold , marked as accidental touch;

[0076] If C freq ≤C threshold , marked as normal click;

[0077] See also Figure 5 The flowchart of the method for determining abnormal click position is shown, and the method for determining abnormal click position is as follows: a page edge area threshold and a distance threshold are set, and if the position coordinates clicked by the user are within the page edge area threshold and the distance from the product display area is greater than the distance threshold, it is marked as an accidental touch; if the position coordinates clicked by the user are outside the page edge area threshold and the distance from the product display area is less than and / or equal to the distance threshold, it is marked as a normal click;

[0078] Specifically, the page edge area threshold R is set edgeThe distance from the edge of the page is 2 mm, and the distance threshold D threshold is 8mm; compare the position coordinates clicked by the user with the page edge area threshold R edge , and the distance d between the user's clicked location coordinates and the product display area and the distance threshold D threshold Size:

[0079] If (x click ,y click ) in R edge In and d>D threshold , marked as accidental touch;

[0080] If (x click ,y click ) is not in R edge Inner or d≤D threshold , marked as normal click;

[0081] See also Figure 6 The flowchart of the method for determining whether a click is inconsistent with a historical behavior is shown, and the method for determining whether a click is inconsistent with a historical behavior includes: setting a matching threshold to match the category of a product clicked by a user with the category of the product browsed by the user in the past; if the matching degree between the type of the product clicked by the user and the category of the product browsed by the user in the past is less than the matching threshold, marking it as an accidental touch; if the matching degree between the type of the product clicked by the user and the category of the product browsed by the user in the past is greater than and / or equal to the matching threshold, marking it as a normal click;

[0082] Specifically, set the matching threshold M threshold is 0.8, comparing the matching degree between the product category clicked by the user and the product category in the user's browsing history:

[0083] If M<M threshold , marked as accidental touch;

[0084] If M ≥ M threshold , marked as a normal click.

[0085] In the embodiment of the present application, by setting the false touch judgment conditions and marking the false touch behavior, the false touch behavior marking data is obtained; by using the dwell time as an intuitive indicator, the click behavior caused by the wrong operation is quickly identified; by setting the frequent click judgment, the behavior of sliding false touch or repeated clicks due to page freeze is accurately captured; by identifying the degree of deviation between the click area and the product display, the false touch behavior of the user who may accidentally touch the edge area of ​​the page or click outside the product area due to slipping when browsing the page is captured; by combining the user's historical interest modeling, the abnormal click data is filtered out, and the correlation between the user's interest data and the recommendation results is strengthened. By marking the user's behavior by false touch, the marked false touch data can be used as the input data of the subsequent false touch detection model of the e-commerce platform, helping the false touch detection model of the e-commerce platform to identify complex false touch patterns and accurately identify the user's false touch behavior.

[0086] Further, by Figure 2 The mis-touch behavior detection module in the system shown builds an e-commerce platform mis-touch detection model, and uses the LSTM model to perform mis-touch detection on the user's click behavior according to the mis-touch features and the mis-touch behavior annotation data to obtain the mis-touch probability; the e-commerce platform mis-touch detection model is built by the LSTM model, and the LSTM model includes: an input layer, an LSTM layer, a mis-touch detection layer, and a mis-touch probability output layer;

[0087] The input layer includes: combining the user behavior data, the time step and the feature dimension of the false touch feature to form an input time series matrix;

[0088] The LSTM layer includes: extracting the behavior pattern in the time series through the LSTM model, processing the long-term dependency information required for false touch detection, and obtaining the hidden state; the specific calculation process includes: input gate, forget gate, output gate, memory unit update and hidden state update;

[0089] The formula of the input gate is:

[0090] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0091] Among them, i t is the control value of the input gate; W i is the weight matrix of the input gate; b i is the bias of the input gate; σ is the sigmoid activation function; h t-1 is the hidden state of the previous time step; x t is the current input; t is the current time step;

[0092] The formula of the forget gate is:

[0093] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0094] Among them, f t Represented as the output of the forget gate; W f is the weight matrix of the forget gate; b f is the bias of the forget gate;

[0095] The formula for the output gate is:

[0096] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0097] Among them, t is the output of the output gate; W o is the weight matrix of the output gate; b o is the bias of the output gate;

[0098] The formula for updating the memory unit is:

[0099] c t =f t ·c t-1 +i t tanh(W c ·[h t-1 ,x t ]+b c );

[0100] Among them, c t is the memory unit of the current time step; c t-1 is the memory unit of the previous time step; tanh is the tanh activation function; W c is the weight matrix of the memory unit; b c is the bias of the memory unit;

[0101] The formula for the hidden state is:

[0102] h t =o t ·tanh(c t );

[0103] Among them, h t is the hidden state of the current time step, which is used for the output of the LSTM model;

[0104] The false touch detection layer includes: mapping the hidden state obtained by the LSTM layer to the probability value of false touch through a fully connected layer; the specific formula for mapping the hidden state to the false touch probability value by the fully connected layer is:

[0105] z=W D ·h t +b D ;

[0106] Where z is the linear output value without activation; W D is the weight matrix of the fully connected layer; b D is the bias of the fully connected layer; the false touch probability output layer includes: outputting the false touch probability through an activation function; the specific formula is:

[0107]

[0108] Where p is the probability of false touch, p∈[0,1];

[0109] Further, a false touch determination threshold is set, and a false touch determination is performed on the false touch probability according to the false touch determination threshold; specifically, in the embodiment of the present application, the false touch determination threshold is set to 0.8; when p≥0.8, the click is determined to be a false touch; when p<0.8, it is determined to be a normal click;

[0110] The embodiment of the present application constructs an e-commerce platform false touch detection model, which uses the LSTM model to detect false touches on the user's click behavior, and can extract long-term dependency information in the time series and adapt to the user's dynamic click behavior; through multi-feature input and the nonlinear mapping capability of deep learning, the LSTM model can effectively distinguish between normal clicks and false touch behaviors, avoiding misjudgments caused by a single rule; constructing an e-commerce platform false touch detection model through the LSTM model is helpful to process user behavior data in real time, and adaptively adjust the false touch detection results according to the input features, thereby improving the false touch detection capability and accuracy of the e-commerce platform false touch detection model for the user's click behavior.

[0111] Further, by Figure 2 The mis-touch behavior detection module in the system shown trains and optimizes the parameters of the mis-touch detection model of the e-commerce platform, and evaluates the mis-touch detection performance of the mis-touch detection model of the e-commerce platform for user click behavior; the training and optimization of the mis-touch detection model of the e-commerce platform includes: dividing the user behavior data into a training set, a validation set and a test set; specifically, in the embodiment of the present application, the user behavior data is divided into 70% training set, 15% validation set and 15% test set;

[0112] The e-commerce platform mis-touch detection model is trained using the training set; the mis-touch detection loss function is calculated using the validation set; the formula of the loss function is:

[0113]

[0114] Among them, L is the loss function value; N is the number of training data; y i is the true label; p i is the probability of false touch;

[0115] Optimizing the e-commerce platform false touch detection model by adjusting the batch size, time step and optimizer parameters of the e-commerce platform false touch detection model; performing false touch detection on user behavior data using the optimized e-commerce platform false touch detection model;

[0116] Table 1 shows the detection results of the e-commerce platform false touch detection model proposed in the embodiment of the present application for false touch detection of the click behaviors of 5 user IDs randomly selected from the collected user behavior data.

[0117] Table 1 False touch detection results

[0118]

[0119] As can be seen from Table 1, the time that users U729, U639, and U391 stay on the current page is less than 2 seconds, indicating that the users exited immediately after clicking on the product page, and the matching degree between the product type and the browsing history is low, which can be determined as an accidental click; at the same time, user U729 clicked frequently within one minute, and the network type was 4G, which can be determined as an accidental touch caused by poor network and lag; while users U174 and U843 stayed on the current page for a long time, indicating that the users browsed the product page carefully, and the product matching degree was higher than 80%, which can be determined as a normal click.

[0120] Furthermore, the test set is used to evaluate the user click and accidental touch detection performance of the optimized e-commerce platform accidental touch detection model, and the accuracy, recall rate and F1 score of the user click and accidental touch detection of the e-commerce platform accidental touch detection model are calculated.

[0121] Table 2 shows the detection performance of the e-commerce platform accidental touch detection model for the five user click behaviors in Table 1.

[0122] Table 2. Accidental touch detection performance of the e-commerce platform accidental touch detection model

[0123] index Accuracy Recall F1 score Numeric 0.94 0.90 0.92

[0124] Depend on Figure 2The accidental touch behavior processing module in the system processes the user's accidental touch behavior according to the accidental touch detection results of the accidental touch detection model of the e-commerce platform to adjust the product recommendation strategy of the e-commerce platform; when an accidental touch is detected, the product is automatically removed from the recommendation list to avoid affecting subsequent recommendations; specifically, the accidental touch products detected by users U729, users U639 and users U391 are automatically removed from the recommendation list to avoid affecting the recommendation quality of the subsequent product recommendation system; at the same time, more content similar to the currently clicked product is recommended to users U174 and users U843 to improve the accuracy of the recommendation system.

[0125] The embodiment of the present application trains and optimizes the e-commerce platform mistaken touch detection model proposed in the present invention. Through data set division, batch training and parameter tuning, it is ensured that the e-commerce platform mistaken touch detection model has high performance in different scenarios, and the accuracy and real-time performance of the model's mistaken touch identification are improved; when a mistaken touch is detected, the mistakenly touched product is automatically removed from the recommendation list. This automatic removal function of the mistakenly touched product avoids the user's interference with the recommendation list due to mistaken touch behavior, and improves the relevance and accuracy of the recommended content.

[0126] The embodiment of the present application realizes accurate false touch detection of user click behavior by constructing a false touch feature extraction model, setting false touch conditions and constructing an e-commerce platform false touch detection model. First, by collecting user behavior data and integrating various aspects of user behavior data, the user click behavior is analyzed in multiple dimensions to identify the possibility of false touch, providing a reliable data basis for the subsequent construction of a false touch feature extraction module. Then, by constructing a false touch feature extraction model, multi-dimensional false touch features are extracted from user behavior data; through the false touch feature extraction model, the system of the present invention can accurately capture possible false touch patterns and provide high-quality input for subsequent false touch detection and product recommendation system optimization. Secondly, by setting false touch judgment conditions and marking false touch behaviors, false touch behavior marking data is obtained; by marking false touches on user behaviors, the marked false touch data can be used as input data for the subsequent false touch detection model of the e-commerce platform, helping the false touch detection model of the e-commerce platform to identify complex false touch patterns and accurately identify the false touch behavior of users. Then, by building an e-commerce platform false touch detection model, the LSTM model is used to detect false touches on the user's click behavior; through multi-feature input and the nonlinear mapping ability of deep learning, the LSTM model can effectively distinguish between normal clicks and false touches, improving the false touch detection ability and accuracy of the e-commerce platform false touch detection model for user click behaviors. Finally, through data set division, batch training and parameter tuning, the false touch detection performance of the e-commerce platform false touch detection model for user click behaviors is optimized, ensuring that the e-commerce platform false touch detection model has high performance in different scenarios, and improving the accuracy of false touch recognition of the model.

[0127] Embodiment 2

[0128] The embodiment of the present invention provides an e-commerce platform product recommendation system based on big data, including:

[0129] A user behavior data acquisition module is used to collect and store user behavior data B of users on the e-commerce platform, including: click behavior data, page interaction data, user history data, device and network data, and context data;

[0130] The user behavior data includes: click behavior data, page interaction data, user history data, device and network data and context data; wherein, the click behavior data includes: click time, click product ID, click location coordinates and click page area; the page interaction data includes: page dwell time, scrolling behavior (scroll up and scroll down) and touch events (strength and direction of scrolling gesture); the user history data includes: browsing history, shopping cart, search history and purchase history; the device and network data includes: device type (PC, mobile phone and tablet), operating system (MacOS, Windows and Linuks), browser type (Chrome browser, Microsoft Edge browser and QQ browser), screen resolution (1920×1080 and 1280×720) and network type (4G, 5G and WiFi); the context data includes: the product list generated by the recommendation algorithm.

[0131] An accidental touch feature extraction module is used to construct an accidental touch feature extraction model, extract multiple features related to the accidental touch according to the user behavior data, and obtain the accidental touch features; the accidental touch features include: dwell time features, click frequency features, click position features, click and product matching features, device type features, and network type features;

[0132] Among them, the dwell time feature includes: calculating the dwell time of the user on the product page after clicking the product; the click frequency feature includes: recording the number of times the user clicks on the same and / or adjacent products within the predetermined time; the specific extraction method is: setting the predetermined time to 2 seconds, and counting the number of clicks on the same and / or adjacent products by the user within the predetermined time; the click position feature includes: recording the page position coordinates clicked by the user and its distance from the product display area; the click and product matching feature includes: comparing the user's browsing history with the matching degree of the currently clicked product, and calculating the mismatched click behavior; the device type feature includes: obtaining the device type used by the user for clicking, including PC, mobile phone and tablet, and judging the impact of different device types on accidental touches; the network type feature includes: obtaining the network used by the user for clicking, including Wi-Fi, 4G and 5G, and judging the impact of different network types on accidental touches.

[0133] An error touch condition setting module is used to set the error touch judgment condition and mark the error touch behavior based on the error touch feature to obtain the error touch behavior marking data; the error touch judgment condition includes: short-time stay judgment, frequent click judgment, abnormal click position judgment and click and historical behavior inconsistency judgment;

[0134] The method for determining the short-time stay is as follows: a page stay time threshold is set, and if the user stays on the product page for less than the page stay time threshold after clicking on the product, it is marked as an accidental touch; if the user stays on the product page for greater than and / or equal to the page stay time threshold after clicking on the product, it is marked as a normal click;

[0135] The method for determining frequent clicks is: setting a click count threshold, if the number of times a user clicks on the same and / or adjacent products within a predetermined time is greater than the click count threshold, it is marked as an accidental touch; if the number of times a user clicks on the same and / or adjacent products within a predetermined time is less than and / or equal to the click count threshold, it is marked as a normal click;

[0136] The method for determining abnormal click position is as follows: setting a page edge area threshold and a distance threshold, if the position coordinates clicked by the user are within the page edge area threshold and the distance from the product display area is greater than the distance threshold, it is marked as an accidental touch; if the position coordinates clicked by the user are outside the page edge area threshold and the distance from the product display area is less than and / or equal to the distance threshold, it is marked as a normal click;

[0137] The method for determining whether a click is inconsistent with historical behavior includes: setting a matching threshold to match the category of the product clicked by the user with the category of the products he has browsed in the past; if the matching degree between the type of product clicked by the user and the category of the products he has browsed in the past is less than the matching threshold, marking it as a false touch; if the matching degree between the type of product clicked by the user and the category of the products he has browsed in the past is greater than and / or equal to the matching threshold, marking it as a normal click.

[0138] The mis-touch behavior detection module is used to build an e-commerce platform mis-touch detection model, and performs mis-touch detection on the user's click behavior according to the mis-touch feature to obtain the mis-touch probability; the e-commerce platform mis-touch detection model is built through an LSTM model, and the LSTM model includes: an input layer, an LSTM layer, a mis-touch detection layer, and a mis-touch probability output layer;

[0139] Setting a false touch determination threshold, and performing false touch determination on the false touch probability according to the false touch determination threshold;

[0140] An evaluation module for the false touch detection model is used to train and optimize the parameters of the false touch detection model of the e-commerce platform, and evaluate the false touch detection performance of the false touch detection model of the e-commerce platform for user click behavior; the training and optimization of the false touch detection model of the e-commerce platform includes: dividing the user behavior data into a training set, a validation set and a test set; using the training set to train the false touch detection model of the e-commerce platform; optimizing the false touch detection model of the e-commerce platform by calculating the false touch detection loss function through the validation set and adjusting the batch size, time step and optimizer parameters of the false touch detection model of the e-commerce platform; using the optimized false touch detection model of the e-commerce platform to perform false touch detection on the user behavior data;

[0141] Table 3 shows the detection results of the e-commerce platform false touch detection model proposed in the embodiment of the present application for false touch detection of the click behaviors of 5 user IDs randomly selected from the collected user behavior data B.

[0142] Table 3 False touch detection results

[0143]

[0144] As can be seen from Table 3, the time that users U910 and U485 stay on the current page is less than 2 seconds, indicating that the users exited immediately after clicking on the product page, and the matching degree between the product type and the browsing history is low, which can be determined as an accidental click; although user U492 stays on the product page for a long time, the user clicks frequently within one minute, and the matching degree between the product type and the browsing history is low, which can be determined as an accidental touch caused by PC freeze; users U371 and U262 stay on the current page for a long time, indicating that the users browsed the product page carefully, and the product matching degree is higher than 80%, which can be determined as a normal click.

[0145] Using the test set to evaluate the performance of the optimized e-commerce platform accidental touch detection model, and calculating the accuracy, recall rate and F1 score of the user click accidental touch detection of the e-commerce platform accidental touch detection model;

[0146] Table 4 shows the detection performance of the e-commerce platform accidental touch detection model for the five user click behaviors in Table 1.

[0147] Table 4. Accidental touch detection performance of the e-commerce platform accidental touch detection model

[0148] index Accuracy Recall F1 score Numeric 0.95 0.90 0.91

[0149] The accidental touch behavior processing module processes the user's accidental touch behavior according to the accidental touch detection result of the accidental touch detection model of the e-commerce platform to adjust the product recommendation strategy of the e-commerce platform.

[0150] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A product recommendation method for an e-commerce platform based on big data, characterized in that: include: Collect and store user behavior data on e-commerce platforms, including: click behavior data, page interaction data, user history data, device and network data, and contextual data; Constructing an accidental touch feature extraction model, extracting multiple features related to the accidental touch according to the user behavior data to obtain the accidental touch features; the accidental touch features include: dwell time features, click frequency features, click position features, click and product matching features, device type features and network type features; Based on the accidental touch characteristics, the accidental touch judgment conditions are set and the accidental touch behavior is marked to obtain the accidental touch behavior marking data; the accidental touch judgment conditions include: short-time stay judgment, frequent click judgment, abnormal click position judgment and click and historical behavior inconsistency judgment; Construct an e-commerce platform mis-touch detection model, use the LSTM model to perform mis-touch detection on the user's click behavior according to the mis-touch features and the mis-touch behavior annotation data, and obtain the mis-touch probability; set a mis-touch determination threshold, and perform mis-touch determination on the mis-touch probability according to the mis-touch determination threshold; Training and optimizing the parameters of the e-commerce platform false touch detection model, and evaluating the false touch detection performance of the e-commerce platform false touch detection model on user click behaviors; The user's accidental touch behavior is processed according to the accidental touch detection result of the accidental touch detection model of the e-commerce platform, so as to adjust the product recommendation strategy of the e-commerce platform.

2. The method for recommending products on an e-commerce platform based on big data according to claim 1, characterized in that: The click behavior data includes: click time, click product ID, click location coordinates and click page area; the page interaction data includes: page dwell time, scrolling behavior and touch events; the user history data includes: browsing history, shopping cart, search history and purchase history; the device and network data includes: device type, operating system, browser type, screen resolution and network type; the context data includes: the product list generated by the recommendation algorithm.

3. The method for recommending products on an e-commerce platform based on big data according to claim 1, characterized in that: The dwell time feature includes: calculating the dwell time of the user on the product page after clicking on the product; the click frequency feature includes: recording the number of times the user clicks on the same and / or adjacent products within a predetermined time; the click position feature includes: recording the page position coordinates of the user's click and the distance from the product display area; the click and product matching feature includes: comparing the user's browsing history with the matching degree of the currently clicked product, and calculating the mismatched click behavior; the device type feature includes: obtaining the device type used by the user for clicking, including PC, mobile phone and tablet, and judging the impact of different device types on accidental touches; the network type feature includes: obtaining the network used by the user for clicking, and judging the impact of different network types on accidental touches.

4. The method for recommending products on an e-commerce platform based on big data according to claim 1, characterized in that: The method for determining a short-time stay is as follows: a page stay time threshold is set, and if the user stays on the product page for less than the page stay time threshold after clicking on the product, it is marked as an accidental touch; if the user stays on the product page for greater than and / or equal to the page stay time threshold after clicking on the product, it is marked as a normal click; the method for determining frequent clicks is as follows: a click count threshold is set, and if the number of times a user clicks on the same and / or adjacent products within a predetermined time is greater than the click count threshold, it is marked as an accidental touch; If the number of times a user clicks on the same and / or adjacent products within a predetermined time is less than and / or equal to the click number threshold, it is marked as a normal click; the method for determining the abnormal click position is as follows: a page edge area threshold and a distance threshold are set, and if the position coordinates clicked by the user are within the page edge area threshold and the distance from the product display area is greater than the distance threshold, it is marked as an accidental touch; if the position coordinates clicked by the user are outside the page edge area threshold and the distance from the product display area is less than and / or equal to the distance threshold, it is marked as a normal click; The method for determining whether a click is inconsistent with a historical behavior includes: setting a matching threshold to match the category of the product clicked by the user with the category of the product browsed by the user in the past; If the degree of match between the type of product clicked by the user and the category of the products he / she has browsed in the past is less than the matching degree threshold, it is marked as an accidental touch; if the matching degree is greater than and / or equal to the matching degree threshold, it is marked as a normal click.

5. The method for recommending products on an e-commerce platform based on big data according to claim 1, characterized in that: The e-commerce platform false touch detection model is constructed through the LSTM model, and the LSTM model includes: an input layer, an LSTM layer, a false touch detection layer and a false touch probability output layer; wherein, the input layer includes: combining the user behavior data, the time step and the feature dimension of the false touch feature to form an input time series matrix, and inputting the LSTM layer; the LSTM layer includes: extracting the behavior pattern in the time series through the LSTM model, processing the long-term dependency information required for false touch detection, and obtaining the hidden state; the false touch detection layer includes: mapping the hidden state obtained by the LSTM layer to the probability value of the false touch through the fully connected layer; the false touch probability output layer includes: outputting the false touch probability through the activation function.

6. The method for recommending products on an e-commerce platform based on big data according to claim 1, characterized in that: The training and optimization of the e-commerce platform false touch detection model includes: dividing the user behavior data into a training set, a validation set and a test set; using the training set to train the e-commerce platform false touch detection model; optimizing the e-commerce platform false touch detection model by calculating the false touch detection loss function through the validation set and adjusting the batch size, time step and optimizer parameters of the e-commerce platform false touch detection model; using the optimized e-commerce platform false touch detection model to perform false touch detection on user behavior data; when a false touch is detected, automatically removing the product from the recommendation list to avoid affecting subsequent recommendations; and using the validation set to evaluate the user click false touch detection performance of the e-commerce platform false touch detection model.

7. A product recommendation system for an e-commerce platform based on big data, characterized in that: include: User behavior data acquisition module, used to collect and store user behavior data on the e-commerce platform, including: click behavior data, page interaction data, user history data, device and network data, and context data; An accidental touch feature extraction module is used to construct an accidental touch feature extraction model, extract multiple features related to the accidental touch according to the user behavior data, and obtain the accidental touch features; the accidental touch features include: dwell time features, click frequency features, click position features, click and product matching features, device type features, and network type features; An error touch condition setting module is used to set the error touch judgment condition and mark the error touch behavior based on the error touch feature to obtain the error touch behavior marking data; the error touch judgment condition includes: short-time stay judgment, frequent click judgment, abnormal click position judgment and click and historical behavior inconsistency judgment; The mis-touch behavior detection module is used to build an mis-touch detection model for the e-commerce platform, perform mis-touch detection on the user's click behavior according to the mis-touch characteristics, and obtain the mis-touch probability; set the mis-touch determination threshold to perform mis-touch determination on the mis-touch probability; An error detection model evaluation module is used to train and optimize the parameters of the e-commerce platform error detection model, and evaluate the error detection performance of the e-commerce platform error detection model for user click behaviors; The accidental touch behavior processing module processes the user's accidental touch behavior according to the accidental touch detection result of the accidental touch detection model of the e-commerce platform to adjust the product recommendation strategy of the e-commerce platform.

8. The product recommendation system for an e-commerce platform based on big data according to claim 7, characterized in that: The dwell time feature includes: calculating the dwell time of the user on the product page after clicking on the product; the click frequency feature includes: recording the number of times the user clicks on the same and / or adjacent products within a predetermined time; the click position feature includes: recording the page position coordinates of the user's click and the distance from the product display area; the click and product matching feature includes: comparing the user's browsing history with the matching degree of the currently clicked product, and calculating the mismatched click behavior; the device type feature includes: obtaining the device type used by the user for clicking, including PC, mobile phone and tablet, and judging the impact of different device types on accidental touches; the network type feature includes: obtaining the network used by the user for clicking, and judging the impact of different network types on accidental touches.

9. The product recommendation system for an e-commerce platform based on big data according to claim 7, characterized in that: The method for determining a short-time stay is as follows: a page stay time threshold is set, and if the user stays on the product page for less than the page stay time threshold after clicking on the product, it is marked as an accidental touch; if the user stays on the product page for greater than and / or equal to the page stay time threshold after clicking on the product, it is marked as a normal click; the method for determining frequent clicks is as follows: a click count threshold is set, and if the number of times a user clicks on the same and / or adjacent products within a predetermined time is greater than the click count threshold, it is marked as an accidental touch; If the number of times a user clicks on the same and / or adjacent products within a predetermined time is less than and / or equal to the click number threshold, it is marked as a normal click; the method for determining the abnormal click position is as follows: a page edge area threshold and a distance threshold are set, and if the position coordinates clicked by the user are within the page edge area threshold and the distance from the product display area is greater than the distance threshold, it is marked as an accidental touch; if the position coordinates clicked by the user are outside the page edge area threshold and the distance from the product display area is less than and / or equal to the distance threshold, it is marked as a normal click; The method for determining whether a click is inconsistent with a historical behavior includes: setting a matching threshold to match the category of the product clicked by the user with the category of the product browsed by the user in the past; If the degree of match between the type of product clicked by the user and the category of the products he / she has browsed in the past is less than the matching degree threshold, it is marked as an accidental touch; if the matching degree is greater than and / or equal to the matching degree threshold, it is marked as a normal click.

10. The product recommendation system for an e-commerce platform based on big data according to claim 7, characterized in that: The e-commerce platform false touch detection model is constructed through an LSTM model, and the LSTM model includes: an input layer, an LSTM layer, a false touch detection layer and a false touch probability output layer.

Images