Product recommendation method, system and equipment based on sensor and medium

Abstract

The invention is applicable to the field of product recommendation, and discloses a sensor-based product recommendation method, which comprises the following steps of: collecting a product use data set of a user through a sensor, extracting a plurality of feature vectors, and aligning time sequence data to form user product use behavior features under a unified time scale. Then, the similarity between the users is calculated, a similarity matrix is constructed, strong correlation users are screened through a set threshold value, and a user similarity relation graph is formed; and based on the relation graph, identifying user communities with similar interests and behavior patterns, predicting potential relation strength among the users, and constructing a social network for cooperative use of products. And meanwhile, user portraits are enriched in combination with user behavior data in related fields. And finally, based on the constructed social network and the user portrait, comprehensively considering the user similarity and the product similarity, generating a candidate recommendation list, and adjusting weights of different similarities according to historical behaviors of the user, thereby obtaining a personalized product recommendation result.

Description

Technical Field

[0001] The present invention relates to the field of product recommendation, and in particular to a sensor-based product recommendation method, system, device and medium. Background Art

[0002] In the field of modern e-commerce and smart devices, personalized recommendation systems have become an important means to improve user experience and increase sales. Existing recommendation systems mainly rely on users' historical behavior data and explicit feedback, such as browsing history, purchase history, and user ratings. However, these systems have significant deficiencies in capturing users' real-time needs. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art. To this end, the present invention proposes a sensor-based product recommendation method, system, device and medium.

[0004] In a first aspect, the present application proposes a sensor-based product recommendation method, comprising: obtaining a user's product usage data set based on a sensor;

[0005] Extracting feature vectors of time, location, usage frequency, usage duration, and operation sequence from the product usage dataset, and aligning the time series data to obtain user product usage behavior characteristics at a unified time scale;

[0006] According to the obtained user product usage behavior characteristics, the cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users. By setting the similarity threshold, weakly correlated users are filtered out to obtain a strong correlated user similarity relationship graph;

[0007] Based on the user similarity relationship graph, a user community structure with similar interests and behavior patterns is obtained, and at the same time, the potential relationship strength between users is predicted to build a social network for collaborative use of products;

[0008] Enrich user portraits using user behavior data from related fields;

[0009] Based on the constructed social network and the user portrait, a candidate recommendation list is generated in combination with user similarity and product similarity, and the weights of different similarities are adjusted based on the user's historical behavior to obtain a recommendation result.

[0010] More specifically, in the above technical solution, obtaining a user's product usage data set based on a sensor includes:

[0011] Acquire the original product usage data collected by the sensor, and classify the original product usage data according to the type of the data to obtain numerical data, categorized data, and time series data;

[0012] Processing the noise and outliers of the original product usage data by a data cleaning algorithm, identifying and removing invalid data, and obtaining cleaned product usage data;

[0013] Convert the cleaned data of different types into a unified structured representation, use normalization for numerical data, use one-hot encoding for categorical data, and use sliding time windows to extract features for time series data to obtain a standardized structured feature matrix;

[0014] Based on the structured feature matrix, a clustering algorithm is used to group users, and an optimal number of clusters is determined through a clustering evaluation method to identify product usage characteristics of different users;

[0015] Association rule mining algorithms are used to discover frequent item sets and association rules used by products, and to analyze association patterns between different product functions.

[0016] More specifically, in the above technical solution, based on the product usage data set, feature vectors of time, location, usage frequency, usage duration and operation sequence are extracted, and the time series data are aligned to obtain the user product usage behavior characteristics under a unified time scale, including:

[0017] Acquire a product usage dataset, wherein the product usage dataset includes a timestamp, a geographic location, and an operation type;

[0018] sorting the product usage dataset according to the timestamp;

[0019] Using the sliding time window method, the time series features of user behavior are extracted in each time window to obtain the feature vector of each time window. The time series features include usage frequency, usage duration and operation sequence.

[0020] Calculating the optimal matching path between the time series features of different users, and mapping the time series features of different users to a unified time scale;

[0021] On the aligned time series features, by setting a minimum support threshold, extracting operation patterns with a frequency higher than the preset minimum support threshold in different time periods as frequent pattern features of the user;

[0022] The user's time series features and frequent pattern features are concatenated to obtain the user's product usage behavior features.

[0023] More specifically, in the above technical solution, according to the obtained user product usage behavior characteristics, a cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users, and weakly correlated users are filtered by setting a similarity threshold to obtain a strongly correlated user similarity relationship diagram, including:

[0024] Based on the product usage data set, construct a user-product behavior matrix;

[0025] For the user-product behavior matrix, an exponential decay function is used to assign different weights to the behavior data of different time periods to obtain the user-product behavior matrix after time decay, wherein the weight calculation formula of the exponential decay function is: w(t)=exp(-λ(t_nowt)), wherein w(t) is the weight of time t, t_now is the current time, t is the time when the behavior occurs, and λ is the decay coefficient;

[0026] Based on the user-product behavior matrix after time decay, the similarity between any two users is calculated to obtain a user similarity matrix;

[0027] According to a preset similarity threshold, the user similarity matrix is ​​filtered, user pairs with similarity lower than the threshold are regarded as weakly correlated, and the corresponding similarity data are removed;

[0028] Normalizing the filtered user similarity matrix, and mapping the similarity values ​​to the interval [0, 1] using a maximum-minimum normalization method;

[0029] The normalized similarity matrix is ​​subjected to cluster analysis, and users are divided into different groups according to similarity to form a strongly correlated user similarity relationship graph, wherein if a user point does not belong to any cluster, the user is judged as an outlier and is presented separately in the relationship graph.

[0030] More specifically, in the above technical solution, based on the user similarity relationship graph, a user community structure with similar interests and behavior patterns is obtained, and the potential relationship strength between users is predicted to build a social network for collaborative use of products, including:

[0031] The product usage data set includes browsing, clicking and purchasing;

[0032] Calculate the similarity between users based on the behavior data and construct a user similarity relationship graph;

[0033] Taking the user similarity relationship graph as input, mapping the nodes to a low-dimensional vector space to obtain a vector representation of the user;

[0034] Based on the vector representation of the user, community discovery is performed on the user to obtain a user community structure with similar interests and behavior patterns;

[0035] Within the user community, predict the potential relationship strength between users and obtain the implicit connections between users within the community;

[0036] The user community structure and the implicit connections among users within the community are integrated into a product collaborative use social network to construct a product collaborative use social network.

[0037] More specifically, the above technical solution uses user behavior data in related fields to enrich user portraits, including:

[0038] Obtain user behavior data related to the target field, and filter out relevant field data based on the similarity with the target field;

[0039] Preprocessing the related field data, and fusing the preprocessed related field data with the user behavior data in the target field to obtain expanded user behavior data;

[0040] A user profile is constructed based on the expanded user behavior data.

[0041] More specifically, in the above technical solution, the weights of different similarities are adjusted based on the user's historical behavior to obtain the recommendation results, including:

[0042] Identify user behavior patterns and preference changes in different time periods and situations, and give higher weight to recent historical behaviors based on the time decay effect;

[0043] Based on the adjusted weights, the candidate recommendation list is sorted to obtain a recommendation result.

[0044] In a second aspect, the present application also proposes a sensor-based product recommendation system, comprising:

[0045] The first processing module: obtaining the user's product usage data set based on the sensor;

[0046] The second processing module: extracts the feature vectors of time, location, usage frequency, usage duration and operation sequence according to the product usage data set, and aligns the time series data to obtain the user product usage behavior characteristics under a unified time scale;

[0047] The third processing module: according to the obtained user product usage behavior characteristics, the cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users, and the weakly correlated users are filtered out by setting the similarity threshold to obtain the strongly correlated user similarity relationship diagram;

[0048] Fourth processing module: based on the user similarity relationship graph, obtaining the user community structure with similar interests and behavior patterns, and predicting the potential relationship strength between users, to build a product collaborative use social network;

[0049] The fifth processing module: enriching user portraits with user behavior data in related fields;

[0050] The sixth processing module: based on the constructed social network and the user portrait, combined with user similarity and product similarity, generates a candidate recommendation list, adjusts the weights of different similarities based on user historical behavior, and obtains the recommendation result node to store the original element information.

[0051] In a third aspect, the present application further proposes a terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described above when executing the computer program.

[0052] In a fourth aspect, the present application further proposes a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method as described in any one of the above items is implemented.

[0053] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0054] Through sensors, users’ product usage data sets are collected and analyzed in real time, including time, location, usage frequency, usage duration, and operation sequence. After data cleaning and feature extraction, the sliding time window is used to align the time series data, and the cosine similarity algorithm based on time decay is used to calculate user similarity and construct a strongly correlated user similarity relationship graph, thereby identifying user communities and building a social network for collaborative product use. At the same time, user behavior data in related fields are combined to enrich user portraits. Finally, personalized product recommendation results are generated based on social networks and user portraits, achieving more accurate and real-time user product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0056] Figure 1 is a flow chart of a sensor-based product recommendation method provided by an embodiment of the present invention;

[0057] Figure 2is a structural diagram of a sensor-based product recommendation system provided by an embodiment of the present invention;

[0058] Figure 3 It is a structural diagram of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0060] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0061] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0062] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0063] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0064] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0065] See also Figure 1 , the present application proposes a sensor-based product recommendation method, including: S101 obtaining a user's product usage data set based on a sensor, S102 extracting feature vectors of time, location, usage frequency, usage duration, and operation sequence based on the product usage data set, and aligning the time series data to obtain the user product usage behavior characteristics under a unified time scale, S103 calculating the similarity matrix between users based on the obtained user product usage behavior characteristics using a cosine similarity algorithm based on time decay, filtering weakly correlated users by setting a similarity threshold, and obtaining a strongly correlated user similarity relationship graph, S104 obtaining a user community structure with similar interests and behavior patterns based on the user similarity relationship graph, and predicting the potential relationship strength between users to build a product collaborative use social network, S105 enriching user portraits using user behavior data in related fields; S106 generating a candidate recommendation list based on the constructed social network and the user portrait, combining user similarity and product similarity, adjusting weights of different similarities based on user historical behavior, and obtaining a recommendation result node to store original element information.

[0066] For the original product usage data collected by sensors, the data cleaning algorithm is used to remove noise and outliers, and the data standardization method is used to convert different types of data (such as numerical, categorical, and time series) into a unified structured representation to obtain a standardized product usage data set. Including:

[0067] The original product usage data collected by the sensor is obtained and classified according to the data type to obtain numerical data, categorical data and time series data; the corresponding data cleaning algorithm is used to process the noise and outliers in the original data, and the threshold is set by statistical analysis method to identify and remove invalid data to obtain the cleaned product usage data; according to the data standardization method, the cleaned different types of data are converted into a unified structured representation, and the normalization method is used for numerical data, the unique hot encoding is used for categorical data, and the sliding time window is used for time series data to extract features to obtain a standardized structured feature matrix; for the standardized product usage data set, the clustering algorithm is used to group users, and the optimal number of clusters is determined by the clustering evaluation method to identify the product usage characteristics of different user groups; the association rule mining algorithm is used to discover the frequent item sets and association rules of product usage, and analyze the association patterns between different product functions; the time series prediction algorithm is used to predict the product usage trend in the future based on the historical product usage data; the standardized product usage data set and the analysis results of the machine learning algorithm are visualized to generate a scatter plot of user clustering results, a network diagram of association rules, and a line chart of product usage trends, which intuitively display the key characteristics and laws of product usage.

[0068] Specifically, from the original product usage data collected by the sensor, we can obtain information such as user operation records and device status. By analyzing the data types, we can divide them into numerical data (such as the number of operations, usage time, etc.), categorical data (such as operation type, device model, etc.) and time series data (such as operation timestamp, device status change, etc.). In view of the noise and outliers that may exist in the original data, data cleaning algorithms such as median filtering and outlier detection are used to process them. For example, by statistically analyzing the distribution of the number of user operations, the upper and lower thresholds are set as the mean ± 3 times the standard deviation, and the data beyond the threshold range is regarded as invalid data and removed to obtain the cleaned product usage data. Then, the data standardization method is used to preprocess the cleaned data. For numerical data, the maximum and minimum value normalization is used to map it to the [0,1] interval; for categorical data, it is converted into a vector representation using unique hot encoding; for time series data, a sliding time window is used to extract features, for example, with a window size of 1 hour, the number of operations and average usage time in each time window are extracted, and finally a standardized structured feature matrix is ​​obtained. On this basis, the K-means clustering algorithm is used to group users, and the clustering effect under different cluster numbers is evaluated by the elbow rule. For example, when the number of clusters is 4, the characteristics of various user groups can be summarized as: high-frequency user group, long-time user group, occasional user group and rare user group, thereby identifying the product usage characteristics of different users. Furthermore, the Apriori association rule mining algorithm is used, with the minimum support of 05 and the minimum confidence of 8 as the threshold, to mine the frequent item sets and association rules of product use, and discover the association patterns between different operations, such as "view product details page → add to shopping cart → submit order", etc.

[0069] Association rule mining algorithms are used to discover frequent item sets and association rules used by products, and to analyze the association patterns between different product functions. This can also include: for time series data, a sliding time window method can be used to extract features, such as the number of operations, average usage time, etc.; then, association rule mining algorithms (such as an improved version of Apriori or FP-Growth) are applied to discover frequent item sets and association rules on time series features, and to analyze the association patterns of product functions in different time periods. If the usage patterns of product functions in different time periods are discovered by analyzing time series data, for example, some functions may have peak usage patterns in the morning or evening, while other functions may be more active at noon or in the afternoon; then, based on the peak usage period of the function, the recommendation system can prioritize the display of related products or functions in a specific time period. For example, if certain functions are frequently used in the morning, more products or services related to the function can be recommended in the morning period.

[0070] Optionally, use the LSTM time series prediction algorithm, taking the number of operations and usage time of users in the past month as input, to predict the product usage trend in the next week. Finally, visualize the analysis results through visual charts, such as drawing scatter plots of user groups, network diagrams of association rules, and line charts of product usage trends, to provide data support for product optimization and marketing decisions.

[0071] Optionally, sensors may include: Operation behavior sensors: These sensors may include touch screen sensors, etc., which are used to capture the user's direct operation behaviors, such as clicking, sliding, pressing keys, etc. These operation behaviors are an important part of the user product usage data set. Device status sensors: Sensors built into the device, such as accelerometers, gyroscopes, temperature sensors, humidity sensors, etc., can monitor the status and environmental conditions of the device in real time. These data help to understand the performance of the device during use and the environment in which the user is located. Position and positioning sensors: GPS sensors, Wi-Fi positioning sensors, etc. can provide users' geographic location information, which is very useful for analyzing users' product usage habits in different locations. Sound and image sensors: Sensors such as microphones and cameras can capture sound and image information during the interaction between users and products. For example, users may interact with products through voice commands, or cameras can capture users' gestures and expressions when operating products. Timestamp sensors: Although not physical sensors in the traditional sense, timestamp information is crucial for building user product usage data sets. It records the time point when each operation behavior occurs, which helps to analyze users' usage habits in different time periods. Other specific sensors: Depending on the specific functions and usage scenarios of the product, other specific types of sensors may also be involved. For example, for smart home products, they may include light sensors, door and window sensors, etc.; for sports and fitness products, they may include heart rate sensors, step sensors, etc. Sensors are well-known technologies in the field, and their specific working principles are not described in detail in this application. Those skilled in the art can select them according to actual needs.

[0072] In some embodiments, feature vectors of multiple dimensions, including time, location, usage frequency, usage duration, and operation sequence, are extracted from standardized product usage data sets, and time series data of different granularities and sampling frequencies are aligned using a multi-sequence alignment algorithm to obtain user product usage behavior characteristics under a unified time scale. This includes:

[0073] Acquire a product usage dataset, wherein the product usage dataset includes a timestamp, a geographic location, and an operation type;

[0074] Preprocess the dataset used by the product to remove missing values ​​and outliers, and sort the data according to timestamps;

[0075] Using the sliding time window method, the multi-dimensional features of user behavior are extracted in each time window to obtain the feature vector of each time window. The multi-dimensional features include usage frequency, usage duration, and operation sequence.

[0076] The DTW algorithm in the dynamic time warping algorithm is used to calculate the optimal matching path between different user time series and map the time series features of different users to a unified time scale.

[0077] On the aligned time series features, the FP-Growth algorithm is used to mine the frequent patterns of user operation sequences. By setting the minimum support threshold, the operation patterns with a frequency higher than the preset threshold in different time periods are extracted as the high-level features of the user behavior pattern.

[0078] The extracted user behavior features of multiple dimensions are combined to obtain a comprehensive feature vector of the user's product usage behavior;

[0079] According to the comprehensive feature vector, a user portrait model, a personalized recommendation model and an abnormal behavior detection model are established. The user portrait model is obtained by clustering analysis of user behavior characteristics, the personalized recommendation model is obtained by a collaborative filtering algorithm, and the abnormal behavior detection model is obtained by establishing a baseline model of normal behavior.

[0080] Specifically, when obtaining the product usage data set, the user's operation log is collected through the embedding point, including information on multiple dimensions such as the operation timestamp accurate to milliseconds, geographic location latitude and longitude, and operation type coding. The collected raw data is preprocessed, and missing values ​​and outliers are identified through statistical analysis. For example, data with timestamps beyond a reasonable range, geographic locations beyond national boundaries, and operation types that are not within the established coding are eliminated, and the data is sorted in ascending order according to the timestamp. A sliding time window with a length of 1 hour and a sliding step of 10 minutes is used to extract user behavior features in each window, including the frequency of use, that is, the total number of operations in the time window, the duration of use, that is, the time difference between the first and last operations in the window, and the operation sequence, that is, the coding sequence of the operation type in the window, thereby obtaining the user's multi-dimensional time series features. For the time series features of different users, the DTW algorithm in dynamic time warping is used for alignment. By calculating the distance matrix of the local intervals of different sequences, dynamic programming is used to solve the global optimal matching path, and the time series features of different users are mapped to a unified standard time axis. On the aligned time series features, the FP-Growth algorithm is used to mine the frequent patterns of user operation sequences. The minimum support is set to 20%, and the operation subsequences with a frequency of more than 20% in different time periods such as daily, weekly, and monthly are extracted, and the extracted frequent subsequences are used as high-level features of user behavior patterns. Finally, the user behavior features of multiple dimensions such as time series features and frequent pattern features are spliced ​​to obtain a comprehensive feature vector of user product usage behavior. Based on the comprehensive feature vector, the K-Means clustering algorithm is used to automatically group users. The optimal number of clusters is 5 through the silhouette coefficient evaluation, and the user portrait model is obtained. At the same time, the collaborative filtering algorithm is used to build a personalized recommendation model. The similarity between users is calculated by cosine similarity. The combination of UserCF and ItemCF is used to make association recommendations for users to "watch again and again" and "buy again and again". In addition, the One-ClassSVM algorithm is used to establish a baseline model of normal user behavior, and anomaly detection is performed on the new user operation sequence. If the anomaly score exceeds the preset threshold, it is determined to be an abnormal user and triggers an early warning.

[0081] In some embodiments, based on the acquired user product usage behavior characteristics, a cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users, and weakly correlated user pairs are filtered by setting a similarity threshold to obtain a high-quality user similarity relationship graph. Including:

[0082] Obtain the user's usage behavior data for each product in different time periods to construct a user-product behavior matrix; for the user-product behavior matrix, use an exponential decay function to assign different weights to the behavior data in different time periods to obtain a user-product behavior matrix after time decay, wherein the weight calculation formula of the exponential decay function is: w(t) = exp(-λ(t_nowt)), wherein w(t) is the weight of time t, t_now is the current time, t is the time when the behavior occurs, and λ is the decay coefficient; based on the user-product behavior matrix after time decay, calculate the similarity between any two users through the cosine similarity algorithm to obtain a user similarity matrix; according to a preset similarity threshold, the user similarity is calculated. The matrix is ​​filtered, and user pairs with similarities lower than the threshold are regarded as weakly correlated, and the corresponding similarity data are eliminated; the filtered user similarity matrix is ​​normalized, and the maximum-minimum normalization method is used to map the similarity values ​​to the interval [0, 1]; the normalized similarity matrix is ​​input into the DBSCAN clustering algorithm for clustering analysis, and users are divided into different groups according to the similarity to form a user similarity relationship graph, wherein, if a user point does not belong to any clustering cluster, the user is judged as an outlier and presented separately in the relationship graph; for different products, user similarity matrices and relationship graphs are constructed based on the user behavior data of each product, respectively, to reveal the similarity relationship between users under different products, and provide a refined basis for personalized recommendation.

[0083] Specifically, first obtain the user's usage behavior data for each product in different time periods to construct a user-product behavior matrix. For example, you can collect the frequency and duration of use of 10 products by users in the past month to form a 1000×10 matrix. Then, for this matrix, use the exponential decay function to assign different weights to the behavior data in different time periods to obtain the user-product behavior matrix after time decay; the weight calculation formula of the exponential decay function is: w(t) = exp(-λ(t_now-t)), λ is 0.5; then, based on the matrix after time decay, calculate the similarity between any two users through the cosine similarity algorithm to obtain the user similarity matrix; set the similarity threshold to 6, and consider the user pairs with similarity lower than the threshold as weakly correlated, and remove the corresponding similarity data; perform maximum-minimum normalization on the filtered similarity matrix to map the similarity value to the [0,1] interval; finally, input the normalized similarity matrix into the DBSCAN clustering algorithm, set the radius parameter eps = 5, and the minimum number of included points minPts = 5, and perform user clustering analysis. Outlier users are presented separately in the relationship diagram; constructing user similarity matrices and relationship diagrams for different products can reveal the similar relationships between users in using different products, providing a more refined basis for personalized recommendations. For example, if it is found that users A and B are highly similar in using product P1, but have large differences in using product P2, then user B's usage preference for product P2 can be specifically recommended to user A, thereby improving the accuracy of the recommendation.

[0084] In some embodiments, based on the user similarity relationship graph, a graph embedding algorithm is used to learn the user's low-dimensional vector representation, and a spectral clustering algorithm is used to discover the user community structure. At the same time, a link prediction algorithm is used to predict the potential relationship strength between users, and a dynamically evolving product collaborative use social network is constructed. This includes:

[0085] Obtaining user behavior data on the product, including browsing, clicking, and purchasing;

[0086] Calculate the similarity between users based on the behavior data and construct a user similarity relationship graph;

[0087] Taking the user similarity relationship graph as input, using the graph embedding algorithm DeepWalk to map the nodes into a low-dimensional vector space to obtain a vector representation of the user;

[0088] Based on the vector representation of the user, a K-Means clustering algorithm is used to discover the user community to obtain a user community structure with similar interests and behavior patterns;

[0089] Within the user community, a matrix decomposition algorithm is used to predict the potential relationship strength between users to obtain the implicit connections between users within the community;

[0090] Integrate the user community structure and the implicit connections among users within the community into the product collaborative use social network to construct a dynamically evolving network structure;

[0091] In the dynamically evolving network structure, user similarity and community structure are updated in real time by analyzing interaction behaviors and product usage patterns between users;

[0092] Using vector representations of users in the user community, constructing user profiles;

[0093] Based on the user portrait, personalized product recommendations and social recommendations are provided to users through collaborative filtering algorithms.

[0094] Specifically, we first obtain the user behavior data on the product, including browsing, clicking, and purchasing information. Then, based on the user behavior data, we use the cosine similarity algorithm to calculate the similarity between users and construct a user similarity relationship graph. For example, if user A and user B browse product X 10 times and 8 times, click 5 times and 6 times, and purchase 2 times and 3 times, respectively, we can calculate that the similarity between the two users on product X is 85. Then, we use the user similarity matrix as input, and use the DeepWalk algorithm to embed the user nodes and map the users into a 128-dimensional vector space. In the vector space, we use the K-Means clustering algorithm to discover the user community, and divide the users with a similarity greater than 7 into the same community. Within each user community, we use the non-negative matrix decomposition algorithm to predict the potential relationship strength between users, and obtain the implicit connection between users within the community. For example, although user C and user D do not have direct interaction, they are found to have a higher relationship strength of 6 through matrix decomposition. We integrate the user community structure and the implicit connection between users within the community into the product collaborative use social network to construct a dynamically evolving network structure. When new user behavior data is generated, user similarity and community structure are updated in real time through incremental learning, so that the network structure can adapt to changes in user interests. Finally, optionally, the vector representation of users in the user community is used, combined with user demographic attributes, historical behavior and other information, to build user portraits through a logistic regression model. Based on the user portrait, an item-based collaborative filtering algorithm is used to calculate the user's interest in products that have not been browsed, and the TopN products with the highest interest are selected and recommended to the user. At the same time, the Top5 users with the highest similarity with the user in the user's community are selected as the candidate list for social recommendations. Through the above methods, personalized product recommendations and social recommendation services can be provided to users to improve user satisfaction and stickiness.

[0095] In some embodiments, in order to solve the problem of sparse historical user behavior data and cold start, the user profile is enriched with user behavior data in related fields through transfer learning methods, and a multi-task learning algorithm that jointly optimizes user similarity and product recommendation is used to improve the recommendation accuracy of cold start users and long-tail products. Including:

[0096] Acquire user behavior data related to the target domain, and filter out a subset of related domain data based on the similarity with the target domain; perform data cleaning and normalization on the subset of related domain data to obtain pre-processed related domain data; fuse the pre-processed related domain data with the target domain data to obtain an expanded user behavior data set; based on the expanded user behavior data set, use a collaborative filtering algorithm to build a user portrait to obtain the user's implicit feedback data; use the implicit feedback data to generate a low-dimensional vector representation of the user through matrix decomposition technology; build a multi-task learning model with user ID and product ID as input, user portrait and product feature vector as the middle layer, user similarity and product rating as output, and realize knowledge transfer between tasks by sharing hidden layer parameters; for cold start users, predict their interest scores for different products based on registration information and initial behavior to obtain the initial embedding vector of cold start users; for long-tail products, crawl content features and use TF-IDF to build a product vector representation; use the expanded user behavior data set to train the multi-task learning model to obtain an optimized user similarity matrix and product recommendation list.

[0097] Specifically, first, obtain user behavior data related to the target field (such as clothing), including browsing, clicking, purchasing and other behaviors. Then, use algorithms such as Word2Vec to calculate the similarity between each field and the target field, select fields with similarity greater than 8 as the relevant field data subset, perform data cleaning on the subset, remove records with missing values ​​exceeding 20%, and perform maximum and minimum value normalization on numerical features. The processed relevant field data and target field data are merged in a ratio of 8:2 to obtain an expanded user behavior data set, and a user portrait is constructed based on the expanded user behavior data.

[0098] Optionally, based on the expanded user behavior data, constructing a user portrait includes: further cleaning and sorting the expanded user behavior data, removing duplicate, invalid or abnormal data records, and ensuring the accuracy and reliability of the data; normalizing and standardizing the data so that data from different sources and dimensions can be compared and analyzed on the same scale. Extract key feature information from the user behavior data, such as the user's browsing behavior, click behavior, purchase behavior, etc., as well as the time, place, frequency and other dimensions of these behaviors; use statistical analysis and data mining technology to further refine and quantify these features to form a feature vector that can fully reflect user behavior and preferences. Based on the extracted feature vectors, clustering algorithms (such as K-Means, DBSCAN, etc.) are used to group users, and users with similar behavior patterns and preferences are divided into the same group; by analyzing the characteristics and behavior patterns of different user groups, the differences and changes in user needs can be further understood, providing a basis for personalized recommendations and precision marketing. Based on the results of user segmentation, a corresponding user portrait is constructed for each user group, including information on the user's age, gender, occupation, interests, hobbies, consumption habits, etc. The construction of the user portrait can be combined with the user's registration information, historical behavior data, social relationships and other information to form a comprehensive and three-dimensional user portrait model.

[0099] In some embodiments, based on the constructed social network and the user portrait, combined with user similarity and product similarity, a candidate recommendation list is generated, and weights of different similarities are adjusted based on user historical behaviors to obtain recommendation results, including:

[0100] Based on the constructed social network and user portraits, we first generate a preliminary candidate recommendation list by combining the similarities between users and products. Then, based on the user's historical behavior data, we dynamically adjust the weights of user similarity and product similarity in the recommendation process. The analysis of user historical behavior data includes identifying the user's behavior patterns and preference changes in different time periods and situations, and giving higher weights to more recent historical behaviors considering the time decay effect. Finally, based on the adjusted weights, we sort and optimize the candidate recommendation list to obtain personalized recommendation results for each user.

[0101] For example, first, we built a detailed user profile based on the user's social interaction, purchase record, browsing history and other information on the platform, and formed a social network based on the social relationships of these users. Then, we used the collaborative filtering algorithm to combine the similarity between users and the similarity between products to generate a preliminary candidate recommendation list; for example, if a user has similar purchase history and browsing preferences with another user, and both users have purchased a certain type of product, then this type of product will be added to the user's candidate recommendation list. Then, based on the user's historical behavior data, we dynamically adjusted the weights of user similarity and product similarity in the recommendation process, analyzed the user's behavior patterns and preference changes in different time periods and situations, and found that users are more inclined to buy gift products during holidays, and more concerned about daily necessities on weekdays; at the same time, we also considered the time decay effect, and gave higher weights to recent historical behaviors, because recent behaviors can better reflect users' current needs and preferences. Finally, based on the adjusted weights, the candidate recommendation list was sorted and optimized. The recommended products were re-sorted according to each user's personalized needs and preferences, and the products that best met the user's current needs were placed at the top of the list. In this way, each user can get personalized recommendation results for themselves on the platform, thereby improving shopping convenience and satisfaction.

[0102] See also Figure 2 , the application also proposes a sensor-based product recommendation system, including:

[0103] The first processing module 201: obtaining a user's product usage data set based on a sensor;

[0104] Second processing module 202: extracting feature vectors of time, location, usage frequency, usage duration, and operation sequence according to the product usage data set, and aligning the time series data to obtain user product usage behavior characteristics under a unified time scale;

[0105] The third processing module 203: according to the obtained user product usage behavior characteristics, a cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users, and weakly correlated users are filtered by setting a similarity threshold to obtain a strongly correlated user similarity relationship graph;

[0106] Fourth processing module 204: based on the user similarity relationship graph, obtaining a user community structure with similar interests and behavior patterns, and predicting the potential relationship strength between users, to build a product collaborative use social network;

[0107] Fifth processing module 205: enriching user portraits using user behavior data in related fields;

[0108] The sixth processing module 206: based on the constructed social network and the user portrait, combined with user similarity and product similarity, generates a candidate recommendation list, adjusts the weights of different similarities based on user historical behavior, and obtains the recommendation result node to store the original element information.

[0109] It is understandable that if Figure 1 The contents of the sensor-based product recommendation method embodiment shown in the figure are applicable to the sensor-based product recommendation system embodiment. The functions specifically implemented by the sensor-based product recommendation system embodiment are similar to those in the example shown in the figure. Figure 1 The sensor-based product recommendation method embodiment shown is the same as that shown in FIG. Figure 1 The beneficial effects achieved by the sensor-based product recommendation method embodiment shown are also the same.

[0110] It should be noted that the information interaction, execution process and other contents between the above-mentioned systems are based on the same concept as the embodiment of the method of the present invention. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0111] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0112] See also Figure 3 The embodiment of the present invention further provides a computer device 3, comprising: a memory 302 and a processor 301 and a computer program 303 stored in the memory 302. When the computer program 303 is executed on the processor 301, a sensor-based product recommendation method as described in any one of the above methods is implemented.

[0113] The computer device 3 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The computer device 3 may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will appreciate that Figure 3 It is only an example of computer device 3 and does not constitute a limitation on computer device 3. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components, for example, it may also include input and output devices, network access devices, etc.

[0114] The processor 301 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0115] In some embodiments, the memory 302 may be an internal storage unit of the computer device 3, such as a hard disk or memory of the computer device 3. In other embodiments, the memory 302 may also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the computer device 3. Further, the memory 302 may also include both an internal storage unit of the computer device 3 and an external storage device. The memory 302 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory 302 may also be used to temporarily store data that has been output or is to be output.

[0116] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the sensor-based product recommendation method as described in any one of the above methods is implemented.

[0117] In this embodiment, if the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, the steps of the above-mentioned various method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0118] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A sensor-based product recommendation method, characterized in that: The method comprises: Acquire user product usage data sets based on sensors; Extracting feature vectors of time, location, usage frequency, usage duration, and operation sequence from the product usage dataset, and aligning the time series data to obtain user product usage behavior characteristics at a unified time scale; According to the obtained user product usage behavior characteristics, the cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users. By setting the similarity threshold, weakly correlated users are filtered out to obtain a strong correlated user similarity relationship graph; Based on the user similarity relationship graph, a user community structure with similar interests and behavior patterns is obtained, and at the same time, the potential relationship strength between users is predicted to build a social network for collaborative use of products; Enrich user portraits using user behavior data from related fields; Based on the constructed social network and the user portrait, a candidate recommendation list is generated in combination with user similarity and product similarity, and the weights of different similarities are adjusted based on user historical behavior to obtain a recommendation result.

2. The sensor-based product recommendation method according to claim 1, characterized in that: Obtain user product usage data sets based on sensors, including: Acquire the original product usage data collected by the sensor, and classify the original product usage data according to the type of the data to obtain numerical data, categorized data, and time series data; Processing the noise and outliers of the original product usage data by a data cleaning algorithm, identifying and removing invalid data, and obtaining cleaned product usage data; Convert the cleaned data of different types into a unified structured representation. For numerical data, normalization is used. For categorical data, unique hot encoding is used. For time series data, sliding time windows are used to extract features to obtain a standardized structured feature matrix. Based on the structured feature matrix, a clustering algorithm is used to group users, and an optimal number of clusters is determined through a clustering evaluation method to identify product usage characteristics of different users; Association rule mining algorithms are used to discover frequent item sets and association rules used by products, and to analyze association patterns between different product functions.

3. The sensor-based product recommendation method according to claim 1, characterized in that: Based on the product usage dataset, feature vectors of time, location, usage frequency, usage duration, and operation sequence are extracted, and the time series data are aligned to obtain user product usage behavior characteristics under a unified time scale, including: Acquire a product usage dataset, wherein the product usage dataset includes a timestamp, a geographic location, and an operation type; sorting the product usage dataset according to the timestamp; Using the sliding time window method, the time series features of user behavior are extracted in each time window to obtain the feature vector of each time window. The time series features include usage frequency, usage duration and operation sequence. Calculating the optimal matching path between the time series features of different users, and mapping the time series features of different users to a unified time scale; On the aligned time series features, by setting a minimum support threshold, extracting operation patterns with a frequency higher than the preset minimum support threshold in different time periods as frequent pattern features of the user; The user's time series features and frequent pattern features are concatenated to obtain the user's product usage behavior features.

4. The sensor-based product recommendation method according to claim 1, characterized in that: According to the obtained user product usage behavior characteristics, the cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users. By setting the similarity threshold to filter out weakly related users, a similarity relationship diagram of strongly related users is obtained, including: Based on the product usage data set, construct a user-product behavior matrix; For the user-product behavior matrix, an exponential decay function is used to assign different weights to the behavior data of different time periods to obtain the user-product behavior matrix after time decay, wherein the weight calculation formula of the exponential decay function is: w(t)=exp(-λ(t_nowt)), wherein w(t) is the weight of time t, t_now is the current time, t is the time when the behavior occurs, and λ is the decay coefficient; Based on the user-product behavior matrix after time decay, the similarity between any two users is calculated to obtain a user similarity matrix; According to a preset similarity threshold, the user similarity matrix is ​​filtered, user pairs with similarity lower than the threshold are regarded as weakly correlated, and the corresponding similarity data are removed; Normalizing the filtered user similarity matrix, and mapping the similarity values ​​to the interval [0, 1] using a maximum-minimum normalization method; The normalized similarity matrix is ​​subjected to cluster analysis, and users are divided into different groups according to similarity to form a strongly correlated user similarity relationship graph, wherein if a user point does not belong to any cluster, the user is judged as an outlier and is presented separately in the relationship graph.

5. The sensor-based product recommendation method according to claim 1, characterized in that: Based on the user similarity relationship graph, the user community structure with similar interests and behavior patterns is obtained, and the potential relationship strength between users is predicted to build a product collaborative use social network, including: The product usage data set includes browsing, clicking and purchasing; Calculate the similarity between users based on the behavior data and construct a user similarity relationship graph; Taking the user similarity relationship graph as input, mapping the nodes to a low-dimensional vector space to obtain a vector representation of the user; Based on the vector representation of the user, community discovery is performed on the user to obtain a user community structure with similar interests and behavior patterns; Within the user community, predict the potential relationship strength between users and obtain the implicit connections between users within the community; The user community structure and the implicit connections among users within the community are integrated into a product collaborative use social network to construct a product collaborative use social network.

6. The sensor-based product recommendation method according to claim 1, characterized in that: Enrich user profiles with user behavior data from related fields, including: Obtain user behavior data related to the target field, and filter out relevant field data based on the similarity with the target field; Preprocessing the related field data, and fusing the preprocessed related field data with the user behavior data in the target field to obtain expanded user behavior data; A user profile is constructed based on the expanded user behavior data.

7. The sensor-based product recommendation method according to claim 1, characterized in that: Adjust the weights of different similarities based on the user's historical behavior to obtain recommendation results, including: Identify user behavior patterns and preference changes in different time periods and situations, and give higher weight to recent historical behaviors based on the time decay effect; Based on the adjusted weights, the candidate recommendation list is sorted to obtain a recommendation result.

8. A sensor-based product recommendation system, characterized in that: include: The first processing module: obtaining the user's product usage data set based on the sensor; The second processing module: extracts the feature vectors of time, location, usage frequency, usage duration and operation sequence according to the product usage data set, and aligns the time series data to obtain the user product usage behavior characteristics under a unified time scale; The third processing module: according to the obtained user product usage behavior characteristics, the cosine similarity algorithm based on time decay is used to calculate the similarity matrix between users, and the weakly correlated users are filtered out by setting the similarity threshold to obtain the strongly correlated user similarity relationship diagram; Fourth processing module: based on the user similarity relationship graph, obtaining the user community structure with similar interests and behavior patterns, and predicting the potential relationship strength between users, to build a social network for collaborative use of products; The fifth processing module: enriching user portraits with user behavior data in related fields; The sixth processing module: based on the constructed social network and the user portrait, combined with user similarity and product similarity, generates a candidate recommendation list, adjusts the weights of different similarities based on user historical behavior, and obtains the recommendation result node to store the original element information.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images