Intelligent recommendation method based on user behavior features

Abstract

An intelligent recommendation method based on user behavior features. The method comprises data collection, data organization, construction of an LR-based CTR model, construction of a CIN-based CTR model, and model selection and recommendation. On the basis of the design idea of a CTR recommendation model, the method takes into account user behavior features such as viewing, purchasing and favoriting, thus making reasonable recommendations on the basis of behaviors of user groups; in addition, massive user data is cleaned and reshaped to filter out valid data, thereby improving the training speed; furthermore, in order to enhance the prediction efficiency and prediction accuracy, a high-order explicit feature interaction terms algorithm is additionally introduced. The method effectively improves the process of conventional recommendation algorithms, thus enhancing the recommendation accuracy.

Description

An intelligent recommendation method based on user behavior characteristics Technical Field

[0001] The present invention belongs to the technical field of big data analysis, and in particular relates to an intelligent recommendation method based on user behavior characteristics. Background Art

[0002] With the rapid development of the internet, cloud computing, and artificial intelligence in recent years, intelligent recommendation technology has become widely integrated into our daily lives and holds enormous commercial potential. Consequently, major internet companies are attaching great importance to this technology and have conducted extensive research. In this rapidly evolving industry, existing intelligent recommendation systems can readily analyze massive amounts of user data to provide recommendations that best suit each user's behavior based on their personality, behavioral habits, and other factors.

[0003] Researchers have gone through multiple stages of improvement in intelligent recommendation systems, each marked by significant advancements and developments. Early on, collaborative filtering models were often used to construct recommendation algorithms. Later, with the rise of machine learning models, attempts were made to apply machine learning to recommendation systems. Ultimately, due to their superior performance, recommendation models based on the CTR (Click Through Rate) algorithm gradually replaced classic machine learning models and became the current mainstream. The CTR recommendation model collects various feature information about users and recommended content, predicts the click-through rate (CTR) of recommended content, and then filters recommendations based on the CTR, thereby achieving intelligent recommendations.

[0004] Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the present invention provides an intelligent recommendation method based on user behavior characteristics, which includes the following main steps:

[0006] Step 1: Collect user data. The user data includes basic information, implicit information, and behavioral information of the user. This step includes:

[0007] Step 1.1, collect basic information of users: when users register, obtain basic descriptive information characteristics of users, including gender, age, education level and other information.

[0008] Step 1.2, collect the user's implicit information: the user completes the necessary assessment process, and the user's implicit information is extracted according to the assessment process, including various assessment ability values, personality tendencies and other information.

[0009] Step 1.3, collect user behavior information: When the user enters the recommendation page, classify and mark the user's behavioral information on different recommended content feedback, including behavior types such as ignoring, viewing, collecting, purchasing, etc., and record the time of the operation.

[0010] Step 2: Reshape the user data. This step includes:

[0011] Step 2.1, data cleaning: Use the average value or special marking method to clean invalid data in the user data.

[0012] Step 2.2, combination construction: According to the requirements of subsequent screening of CTR models, different combinations of the data collected in step 1 are constructed.

[0013] Step 2.3, data preprocessing: perform corresponding encoding conversion operations and embedding processing operations on the categorical feature data and continuous feature data in the data collected in step 1, so that they can be used as input for the CTR model.

[0014] Step 3: Build a CTR model based on LR. Based on the logistic regression algorithm, construct a CTR model. This step includes:

[0015] Step 3.1, input information features: select the user's basic information features, the user's implicit information features, the user's behavioral information features (whether to click on the label information) and product information as model input.

[0016] Step 3.2, combine informative features: combine these features using a first-order linear relationship function.

[0017] Step 3.3, perform training calculation: Based on the actual click result, perform training calculation to calculate cross entropy on the linear calculation result.

[0018] Step 4: Build a CTR model based on CIN. Based on the cross network principle, build a CTR model. This step includes:

[0019] Step 4.1: Input information features. Select the user's basic information features, the user's implicit information features, the user's behavior information features (whether the tag is clicked), and the product information as model input.

[0020] Step 4.2: Construct the first layer vector. Construct the output vector of the first layer of the compressed interaction network CIN. This step includes:

[0021] In step 4.2.1, the embedded input vector is subjected to Hadamard product calculation with itself to obtain multiple two-dimensional matrices.

[0022] In step 4.2.2, multiple two-dimensional matrices are multiplied by the corresponding coefficient matrix to obtain a feature map vector.

[0023] Step 4.2.3: Repeat step 4.2.2 as needed to obtain multiple feature map vectors to form the first layer of the CIN network.

[0024] Step 4.3: Construct multi-layer vectors. The CIN network vector of the previous layer is calculated with the input vector as in step 4.2 to construct the next layer of CIN network.

[0025] Step 4.4: Compress the results. By using a random value pooling method with a certain coefficient, compress the results of each layer of the CIN network to obtain the output of the CIN network part.

[0026] Step 4.5: Joint training. Construct the DNN and linear parts of the CTR model and perform joint training with the CIN part.

[0027] Step 4.6: Calculate cross entropy. Use the sigmoid function to jointly calculate the outputs of the three parts and calculate the cross entropy as the output of the entire model.

[0028] Step 4.7: Regularize the objective function. Regularize the objective function by adding a Batch Normalization layer to the deep neural network to improve generalization. Dropout is used on neurons to prevent overfitting during training, and L2 regularization is applied. Only L2 regularization is used for the embedding layer and CIN network.

[0029] Step 5: Apply the model to filter recommendations. Using the two-layer recommendation model to filter and obtain recommendation results, this step includes:

[0030] Step 5.1: Preliminary screening: Based on the user's selected category, the corresponding recommended content is put into the LR-based CTR model for preliminary screening.

[0031] Step 5.2: Secondary screening. The results of the initial screening are fed into the CTR model based on CIN for final screening. Finally, the results are sorted by output probability, and the top 20 are obtained as the final output.

[0032] The beneficial effect of the present invention is: proposing an intelligent recommendation method based on user behavior characteristics, which can reasonably utilize user basic information characteristics, user implicit information characteristics, and user behavior characteristics, greatly reduce the training cost of recommendation model training, and improve the accuracy of recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] FIG1 shows a flow chart of an intelligent recommendation method based on user behavior characteristics of the present invention;

[0034] FIG2 shows the use process of an intelligent recommendation method based on user behavior characteristics of the present invention;

[0035] FIG3 shows the specific structure of the core algorithm model of the intelligent recommendation method of the present invention;

[0036] FIG4 shows the collection and conversion process of the input data of the model of the present invention;

[0037] FIG5 and FIG6 show the calculation process of the feature map portion of the CIN network portion of the present invention. DETAILED DESCRIPTION

[0038] The preferred embodiments of the present invention are further described below with reference to the accompanying drawings and examples.

[0039] The flowchart shown in FIG1 shows the basic process of the implementation of the present invention, and the network structure diagram shown in FIG2 shows the structure of the entire innovative model of the present invention:

[0040] Step 1: Collect user data. Collect the user's input feature data in the CTR model. This step includes:

[0041] Step 1.1. Collect basic user information. When a user registers, obtain basic user information such as gender, age, and education level.

[0042] Step 1.2. Collect the user's implicit information. The user completes the necessary assessment process and uses the assessment form to extract the user's implicit information from the information obtained during the assessment process, including various assessed ability values, personality tendencies, and other information.

[0043] Step 1.3. Collect user behavior information. When a user enters the recommendation page, categorize and mark the user's behavior towards different recommended content information, including: ignoring, viewing, saving, purchasing, etc., and record the time of the operation, so as to update the user's behavior information in real time.

[0044] Step 2: Reshape the user data. Reshape the data obtained in step 1 to make it conform to the requirements of the following steps.

[0045] Usage requirements. This step includes:

[0046] Step 2.1. Clean the data. Use the average value or special mark method to clean the invalid data collected in step 1.

[0047] Step 2.2. Combination Construction: Based on the requirements of subsequent CTR model screening, different combinations of the data collected in step 1 are constructed.

[0048] Step 2.3. Data preprocessing. The categorical feature data and continuous feature data collected in step 1 are subjected to corresponding encoding conversion and embedding operations respectively so that they can be used as input for the CTR model. This step includes:

[0049] Step 2.3.1. Perform a simple normalization operation on continuous data and use label encoding conversion operation on categorical data.

[0050] Step 2.3.2. Based on the LR model, the categorical data that has undergone the label encoding conversion operation in step 2.3.1 is normalized, and a joint model training is performed with the continuous data that has undergone the simple normalization operation in step 2.3.1.

[0051] Step 2.3.3. Based on the CIN model, use one-hot encoding to convert the categorical data that has undergone the label encoding conversion operation in step 2.3.1 into sparse vector data. Then, perform embedding operations on it and the continuous data that has undergone simple normalization operations in step 2.3.1 to obtain dense vectors respectively. Use these two dense vectors for model training.

[0052] Step 3: Build a CTR model based on LR. Based on the logistic regression algorithm, construct a CTR model. This step includes:

[0053] Step 3.1. Input information features. Select the user's basic information features, the user's implicit information features, the user's behavior information features (whether the tag is clicked), and the product information as model input.

[0054] Step 3.2. Combine informative features. Combine these features using a first-order linear relationship function.

[0055] Step 3.3. Perform training calculations. Based on the actual click results, perform cross entropy training calculations on the linear calculation results.

[0056] Step 4: Build a CTR model based on CIN. Based on the cross network principle, build a CTR model. This step includes:

[0057] Step 4.1. Input information features. Select the user's basic information features, the user's implicit information features, the user's behavior information features (whether the tag is clicked), and the product information as model input.

[0058] Step 4.2. Construct the first layer vector. Construct the output vector of the first layer of the compressed interaction network CIN. This step includes:

[0059] Step 4.2.1. Calculate the Hadamard product of the embedded input vector with itself to obtain multiple two-dimensional matrices.

[0060] Step 4.2.2. Multiply multiple matrices by the corresponding coefficient matrix to obtain a feature map vector.

[0061] Step 4.2.3. Repeat step 4.2.2 as needed to obtain multiple feature map vectors to form the first layer of the CIN network. The calculation process is shown in Figures 5 and 6. The calculation formula is as follows:

[0062] Step 4.3. Construct multi-layer vectors. The CIN network vector of the previous layer is calculated with the input vector as in step 4.2 to construct the next layer of CIN network. The calculation formula is as follows:

[0063] Step 4.4. Result compression: By using a random value pooling method according to a certain coefficient, the result of each layer of the CIN network is compressed to obtain the output of the CIN network part.

[0064] Step 4.5. Joint training: Construct the DNN part and linear part of the CTR model and jointly train them with the CIN part.

[0065] Step 4.6. Cross entropy calculation. Use the sigmoid function to jointly calculate the outputs of the three parts and perform cross entropy calculation as the output of the entire model. The calculation formula is as follows:

[0066] Step 4.7. Regularize the objective function. Regularize the objective function by adding a batch normalization layer to the deep neural network to improve generalization, using dropout on neurons to prevent overfitting, and applying L2 regularization. Only L2 regularization is used for the embedding layer and CIN network.

[0067] Step 5: Apply the model to filter recommendations. Using the two-layer recommendation model to filter and obtain recommendation results, this step includes:

[0068] Step 5.1. Preliminary screening: Based on the user's selected category, the corresponding recommended content is put into the LR-based CTR model for preliminary screening.

[0069] Step 5.2. Secondary Screening: The results of the initial screening are fed into the CTR model based on CIN for final screening. The results are then sorted by probability, and the top 20 are taken as the final output.

Claims

1. An intelligent recommendation method based on user behavior characteristics, characterized in that, it includes the following steps: Step 1, collect user data; the user data includes the user's basic information, implicit information, and behavior information; Step 2, reshape the user data; Step 3, construct a CTR model based on LR; construct a CTR model based on the logistic regression algorithm; Step 4, construct a CTR model based on CIN; Based on the principle of cross network, construct a CTR model; Step 5, apply the model to screen and recommend; use a two-layer recommendation model to screen and obtain the recommendation results.

2. An intelligent recommendation method based on user behavior characteristics, characterized in that, in step 3, construct a CTR model based on LR; construct a CTR model based on the logistic regression algorithm; Step 3 further includes: Step 3.1, input information features: select the click or not label information and product information in the user's basic information features, user's implicit information features, and user's behavior information features as the model input; Step 3.2, combine information features: use a first-order linear relationship function to combine these features; Step 3.3, perform training calculations: according to the actual click or not result, perform training calculations of cross entropy on the linear calculation results.

3. An intelligent recommendation method based on user behavior characteristics, characterized in that, in step 4, construct a CTR model based on CIN; Based on the principle of cross network, construct a CTR model; Step 4 further includes: Step 4.1, input information features; select the click or not label information and product information in the user's basic information features, user's implicit information features, and user's behavior information features as the model input; Step 4.2, construct the first-layer vector; construct the output vector of the first-layer compressed interaction network CIN. This step includes: Step 4.2.1, multiply the embedded input vector by itself through Hadamard product calculation to obtain multiple two-dimensional matrices; Step 4.2.2, multiply multiple two-dimensional matrices by the corresponding coefficient matrices to obtain a feature map vector; Step 4.2.3, repeat step 4.2.2 as needed to obtain multiple feature map vectors, constituting the first-layer CIN network; Step 4.3, construct multi-layer vectors; perform the same calculations as in step 4.2 on the CIN network vector of the previous layer and the input vector, and continue to construct the next-layer CIN network; Step 4.4, compress the results; compress the results of each layer of the CIN network by using the method of randomly taking values for pooling according to a certain coefficient to obtain the output of the CIN network part; Step 4.5, joint training; construct the DNN part and the linear part in the CTR model, and perform joint training with the CIN part; Step 4.6, calculate cross entropy; use the sigmoid function to jointly calculate the outputs of the three parts and perform cross entropy calculation as the output of the entire model; Step 4.7, Regularize the objective function; Regularize the objective function. Add a Batch Normalization layer to the deep neural network part to improve the generalization ability, use Dropout for neurons to prevent overfitting during training, and adopt L2 regularization; while only L2 regularization is adopted for the Embedding layer and the CIN network part.

4. An intelligent recommendation method based on user behavior characteristics, characterized in that, in step 5, a model is applied for screening and recommendation; The recommendation results are obtained by screening with a two-layer recommendation model; The step 5 further includes: Step 5.1, Preliminary screening; Through the category selected by the user, put the corresponding recommended content into the CTR model based on LR for preliminary screening; Step 5.2, Secondary screening; Send the results after preliminary screening into the CTR model based on CIN for final screening, and finally sort according to the output probability, and take the top 20 results as the final output.

Images