Information recommendation method, electronic device, storage medium and computer program product

Abstract

The invention provides an information recommendation method, electronic equipment, a storage medium and a computer program product. The method comprises the following steps: acquiring historical behavior data of a target user, the historical behavior data comprising user attribute information and historical transaction information associated with the target user; preprocessing the historical behavior data to obtain a preprocessing result; the preprocessing result is analyzed and processed through a target prediction network model, a user intention prediction result is obtained, and the target prediction network model is obtained by conducting machine learning training through multiple sets of sample behavior data; performing dimension reduction processing on the user intention prediction result to obtain a dimension reduction result; target recommendation information is generated based on the dimension reduction result, and the target recommendation information is used for representing the shipping space recommendation content and the feature influence parameters corresponding to the target user. The information recommendation method and device solve the technical problems of low success rate and poor real-time performance of an information recommendation method provided in related technologies.

Description

Technical Field

[0001] The present disclosure relates to the field of information technology, and in particular to an information recommendation method, electronic device, storage medium, and computer program product. Background Art

[0002] In the modern aviation service industry, the recommendation and sale of additional services for passengers plays a vital role in improving the profits and competitiveness of airlines. In the relevant technologies, rule-based recommendation models and statistical model-based recommendation models are usually used to recommend additional services for passengers. The rule-based recommendation model recommends additional services to passengers through preset business rules and logic, while the statistical model-based recommendation model uses historical data and statistical methods to predict and recommend passenger needs. However, the information recommendation method provided in the relevant technology has technical problems such as low success rate and poor real-time performance, which makes it difficult to meet the needs of airlines for personalized, efficient and real-time passenger service recommendations in a rapidly changing market environment.

[0003] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0004] The embodiments of the present disclosure provide an information recommendation method, an electronic device, a storage medium, and a computer program product to at least solve the technical problems of low success rate and poor real-time performance of the information recommendation method provided in the related art.

[0005] According to one aspect of an embodiment of the present disclosure, there is provided an information recommendation method, comprising: obtaining historical behavior data of a target user, wherein the historical behavior data comprises: user attribute information and historical transaction information associated with the target user; preprocessing the historical behavior data to obtain a preprocessing result; analyzing and processing the preprocessing result using a target prediction network model to obtain a user willingness prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data; performing dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result; generating target recommendation information based on the dimensionality reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0006] Optionally, preprocessing the historical behavior data to obtain preprocessing results includes: performing data cleaning processing on the historical behavior data to obtain data cleaning results, wherein the data cleaning processing includes at least one of the following: missing value processing and outlier processing; normalizing the data cleaning results according to a preset feature range to obtain a normalized result; performing feature extraction based on the normalized result to obtain a feature extraction result; filtering the feature extraction results using preset filtering parameters to obtain a feature filtering result, wherein the preset filtering parameters include at least one of the following: correlation analysis parameters or importance assessment parameters; segmenting the feature filtering results to obtain a preprocessing result.

[0007] Optionally, the target prediction network model includes: an encoding module and an analysis module. The target prediction network model is used to analyze and process the preprocessing results to obtain the user willingness prediction results, including: using the encoding module to convert the preprocessing results to obtain a first feature vector; using the analysis module to analyze and process the first feature vector to obtain the user willingness prediction result.

[0008] Optionally, the analysis module includes: a linear layer and a normalized exponential function layer. The analysis module is used to analyze and process the first eigenvector to obtain a user willingness prediction result, including: using the linear layer to perform a linear transformation on the first eigenvector to obtain a first transformation result; using the normalized exponential function layer to perform probability calculation on the first transformation result to obtain a probability distribution result; performing classification processing based on the probability distribution result to obtain a classification result; and determining the user willingness prediction result based on the classification result.

[0009] Optionally, the user willingness prediction result is subjected to dimensionality reduction processing to obtain the dimensionality reduction result, including: converting the user willingness prediction result to obtain a second eigenvector; linearly transforming the second eigenvector to obtain a second transformation result; splicing the second transformation result to obtain a feature splicing result; compressing the feature splicing result to obtain a dimensionality reduction result.

[0010] Optionally, generating target recommendation information based on the dimensionality reduction result includes: performing attention calculation based on the dimensionality reduction result to obtain an attention calculation result; performing residual normalization calculation based on the attention calculation result to obtain a first calculation result; inputting the first calculation result into the target deep neural network model for analysis and processing to generate target recommendation information.

[0011] Optionally, the target deep neural network model includes: multiple convolution layers, rectified linear unit layers, and fully connected layers. The first calculation result is input into the target deep neural network model for analysis and processing. The generation of target recommendation information includes: using the first convolution layer to perform convolution processing on the first calculation result to obtain a first convolution processing result; using the rectified linear unit layer to perform screening processing on the first convolution processing result to obtain a target screening result; using the second convolution layer to perform convolution processing on the target screening result to obtain a second convolution processing result; performing residual normalization processing based on the second convolution processing result to obtain a second calculation result; using the fully connected layer to perform prediction processing on the second calculation result to obtain a target prediction result; and generating target recommendation information based on the target prediction result.

[0012] According to another aspect of an embodiment of the present disclosure, an information recommendation device is also provided, including: an acquisition module, used to acquire historical behavior data of a target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user; a preprocessing module, used to preprocess the historical behavior data to obtain a preprocessing result; a first processing module, used to analyze and process the preprocessing result using a target prediction network model to obtain a user willingness prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data; a second processing module, used to perform dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result; a generation module, used to generate target recommendation information based on the dimensionality reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0013] Optionally, the preprocessing module is also used to: perform data cleaning on the historical behavior data to obtain data cleaning results, wherein the data cleaning includes at least one of the following: missing value processing and outlier processing; normalize the data cleaning results according to a preset feature range to obtain a normalized result; perform feature extraction based on the normalized result to obtain a feature extraction result; filter the feature extraction results using preset filtering parameters to obtain a feature filtering result, wherein the preset filtering parameters include at least one of the following: correlation analysis parameters or importance assessment parameters; segment the feature filtering results to obtain a preprocessing result.

[0014] Optionally, the first processing module is further used to: utilize the encoding module to convert the preprocessing result to obtain a first feature vector; utilize the analysis module to analyze the first feature vector to obtain a user willingness prediction result.

[0015] Optionally, the first processing module is also used to: use a linear layer to perform a linear transformation on the first eigenvector to obtain a first transformation result; use a normalized exponential function layer to perform probability calculation on the first transformation result to obtain a probability distribution result; perform classification processing based on the probability distribution result to obtain a classification result; and determine a user intention prediction result based on the classification result.

[0016] Optionally, the second processing module is also used to: convert the user intention prediction result to obtain a second eigenvector; perform linear transformation on the second eigenvector to obtain a second transformation result; perform splicing on the second transformation result to obtain a feature splicing result; perform compression on the feature splicing result to obtain a dimensionality reduction result.

[0017] Optionally, the generation module is also used to: perform attention calculation based on the dimensionality reduction result to obtain an attention calculation result; perform residual normalization calculation based on the attention calculation result to obtain a first calculation result; input the first calculation result into the target deep neural network model for analysis and processing to generate target recommendation information.

[0018] Optionally, the generation module is also used to: use the first convolution layer to convolve the first calculation result to obtain a first convolution processing result; use the rectified linear unit layer to filter the first convolution processing result to obtain a target filtering result; use the second convolution layer to convolve the target filtering result to obtain a second convolution processing result; perform residual normalization according to the second convolution processing result to obtain a second calculation result; use the fully connected layer to predict the second calculation result to obtain a target prediction result; and generate target recommendation information based on the target prediction result.

[0019] According to another aspect of an embodiment of the present disclosure, an electronic device is further provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the information recommendation method in the embodiment of the present disclosure.

[0020] According to another aspect of an embodiment of the present disclosure, a computer-readable storage medium is further provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program runs, the device where the storage medium is located is controlled to execute the information recommendation method in the embodiment of the present disclosure.

[0021] According to another aspect of an embodiment of the present disclosure, a computer program product is further provided. The computer program product includes computer instructions. When the computer instructions are executed by a processor, the information recommendation method in the embodiment of the present disclosure is implemented.

[0022] In the disclosed embodiment, historical behavior data of target users is obtained and preprocessed to obtain preprocessing results, and then the preprocessing results are analyzed and processed using a target prediction network model to obtain user willingness prediction results, and finally the user willingness prediction results are subjected to dimensionality reduction processing to obtain dimensionality reduction results, and target recommendation information is generated based on the dimensionality reduction results, thereby achieving the purpose of accurately predicting users' preferences for purchasing additional services of airlines, thereby realizing the technical effect of improving the success rate and real-time response speed of the recommendation system, and thereby solving the technical problems of low success rate and poor real-time performance of the information recommendation methods provided in the relevant technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings described herein are used to provide a further understanding of the present disclosure and constitute a part of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure. In the drawings:

[0024] Figure 1 is a flow chart of an information recommendation method according to one embodiment of the present disclosure;

[0025] Figure 2 is a schematic diagram of an information recommendation model according to one embodiment of the present disclosure;

[0026] Figure 3 It is a structural block diagram of an information recommendation device according to one embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the scheme of the present disclosure, the technical scheme in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present disclosure.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0029] In the related art, rule-based recommendation mode and statistical model-based recommendation mode are usually used to recommend additional services for passengers. The rule-based recommendation mode recommends additional services to passengers through preset business rules and logic, while the statistical model-based recommendation mode predicts and recommends passenger needs using historical data and statistical methods. However, the information recommendation method provided in the related art has technical problems such as low success rate and poor real-time performance, which makes it difficult to meet the needs of airlines for personalized, efficient and real-time passenger service recommendations in a rapidly changing market environment.

[0030] Specifically, the rule-based recommendation model uses preset business rules and logic to recommend additional services to passengers, such as uniformly recommending paid upgrade services to passengers on specific flights and routes based on flight route rules. However, this model ignores the diversity of passengers' personalized needs and behavioral characteristics. Its recommendation strategy is too static and cannot dynamically adapt to the specific preferences of each passenger. The recommendations lack pertinence and are difficult to meet the personalized needs of passengers. The recommendation model based on statistical models predicts and recommends services by analyzing passenger historical data and using statistical methods. Based on the passenger's past behavior patterns, for example, the frequency of a passenger purchasing paid seats is counted, and the seat type with the most historical purchases is used as the basis for future recommendations. Compared with the rule-based recommendation model, the recommendation model based on statistical models can capture a certain degree of passenger behavior, thereby improving the accuracy of recommendations. However, in the face of massive real-time updated data in the aviation industry, the response speed and prediction accuracy of the recommendation model based on statistical models are often not ideal, making it difficult to provide instant and accurate personalized recommendation services.

[0031] According to an embodiment of the present disclosure, a method embodiment of an information recommendation method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] The method embodiment can be executed in an electronic device or a similar computing device including a memory and a processor. Taking running on a computer terminal as an example, the computer terminal may include one or more processors (processors may include but are not limited to central processing units (CPU), graphics processing units (GPU), digital signal processing (DSP) chips, microprocessors (MCU), programmable logic devices (Field Programmable Gate Array, FPGA), neural network processors (Neural-network Processor Unit, NPU), tensor processors (TensorProcessing Unit, TPU), artificial intelligence (Artificial Intelligence, AI) type processors and other processing devices) and a memory for storing data. Optionally, the above-mentioned computer terminal may also include a transmission device, an input and output device, and a display device for communication functions. It can be understood by those of ordinary skill in the art that the above-mentioned structural description is only for illustration, and it does not limit the structure of the above-mentioned computer terminal. For example, the computer terminal may also include more or fewer components than the above-mentioned structural description, or have a configuration different from the above-mentioned structural description.

[0033] The memory can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the information recommendation method in the embodiment of the present disclosure. The processor executes various functional applications and data processing by running the computer program stored in the memory, that is, realizing the above-mentioned information recommendation method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the mobile terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0034] The transmission device is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the mobile terminal. In one example, the transmission device includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.

[0035] The display device may be, for example, a touch screen type liquid crystal display (LCD) and a touch display (also referred to as a "touch screen" or "touch display screen"). The liquid crystal display may enable a user to interact with a user interface of a mobile terminal. In some embodiments, the mobile terminal has a graphical user interface (GUI), and a user may interact with the GUI by finger contacts and / or gestures on a touch-sensitive surface, wherein the human-computer interaction functions here may optionally include the following interactions: creating web pages, drawing, word processing, making electronic documents, games, video conferencing, instant messaging, sending and receiving emails, call interfaces, playing digital videos, playing digital music and / or web browsing, etc. The executable instructions for executing the above human-computer interaction functions are configured / stored in a computer program product or readable storage medium executable by one or more processors.

[0036] Figure 1 is a flow chart of an information recommendation method according to one embodiment of the present disclosure. Figure 1 As shown, the method comprises the following steps:

[0037] Step S11, obtaining historical behavior data of the target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user;

[0038] Step S12, preprocessing the historical behavior data to obtain a preprocessing result;

[0039] Step S13, analyzing and processing the preprocessing results using a target prediction network model to obtain a user intention prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data;

[0040] Step S14, performing dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result;

[0041] Step S15, generating target recommendation information based on the dimension reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0042] The above-mentioned user attribute information refers to a series of static data that can reflect the personal characteristics and travel preferences of the target user, including but not limited to basic personal information such as the user's age and gender, and specific attributes closely related to air travel, such as the user's frequent flyer membership status in a certain airline, membership level, accumulated flight mileage or points, and whether there are specific travel preferences or needs.

[0043] The above-mentioned historical transaction information refers to a detailed record of transactions between the target user and the airline during past air travel, including ticket purchase history (such as purchase time, fare, purchase channel, etc.), seat preference history (such as specific seat selection, seat upgrade, etc.), baggage service history (such as additional baggage purchase, baggage weight level, etc.), in-flight service consumption records (such as selection and use of catering and entertainment services) and purchase information of other additional services (such as priority boarding, travel insurance, etc.).

[0044] The target prediction network model mentioned above refers to a deep learning model used to predict the target user's willingness to purchase additional services. Common target prediction network models include: Deep Neural Network (DNN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory Network (LSTM), Gated Recurrent Unit (GRU), Residual Network (ResNet), Attention Mechanism Integration Model, Graph Neural Network (GNN), etc.

[0045] Exemplarily, firstly, the historical behavior data of the target user is collected and sorted, including ticket purchase records, seat selection preferences, upgrade history, additional baggage purchase records, and basic user information (such as frequent flyer card information, frequent flyer level, purchase history), etc., and the above data is preprocessed to obtain the preprocessing result; secondly, the preprocessing result is input into the residual network for analysis and processing to predict the user's willingness to purchase specific additional services, and the user willingness prediction result is reduced in dimension using representation concatenation and representation compression to obtain the reduced in dimension processing result; finally, the cabin recommendation content and feature influencing parameters corresponding to the target user are generated based on the obtained reduced in dimension processing result.

[0046] Based on the above steps S11 to S15, by acquiring the historical behavior data of the target user and preprocessing the historical behavior data to obtain the preprocessing results, and then using the target prediction network model to analyze and process the preprocessing results to obtain the user intention prediction results, finally the user intention prediction results are subjected to dimensionality reduction processing to obtain the dimensionality reduction results, and target recommendation information is generated based on the dimensionality reduction results, thereby achieving the purpose of accurately predicting the user's purchase preference for additional services of the airline, thereby achieving the technical effect of improving the success rate and real-time response speed of the recommendation system, and thus solving the technical problems of low success rate and poor real-time performance of the information recommendation method provided in the relevant technology.

[0047] Optionally, in step S12, the historical behavior data is preprocessed, and the preprocessing results include:

[0048] Step S121, performing data cleaning processing on the historical behavior data to obtain a data cleaning result, wherein the data cleaning processing includes at least one of the following: missing value processing and outlier processing;

[0049] Step S122, normalizing the data cleaning result according to a preset feature range to obtain a normalized result;

[0050] Step S123, performing feature extraction based on the normalization result to obtain a feature extraction result;

[0051] Step S124, screening the feature extraction result using preset screening parameters to obtain feature screening results, wherein the preset screening parameters include at least one of the following: a correlation analysis parameter or an importance assessment parameter;

[0052] Step S125, segmenting the feature screening result to obtain a preprocessing result.

[0053] The above-mentioned missing value processing refers to a series of operations performed on the missing data in the data set during the data preprocessing process to solve the problem of incomplete data. Missing value processing usually includes: deletion method, filling method, prediction method, etc. Among them, the filling method refers to the use of statistical methods (such as mean, median, mode, 0) or machine learning model prediction results to fill the missing values; the prediction method refers to predicting missing values ​​based on data features by building a data prediction model.

[0054] The above-mentioned outlier processing refers to the process of identifying and processing data points that deviate from the normal range or pattern in the data preprocessing process. Outliers may be caused by data entry errors, measurement errors or extreme events. The processing methods include deleting outliers, correcting outliers, or using statistical thresholds and machine learning methods to identify and adjust outliers to reduce the interference of outliers on model training. Among them, correcting outliers means adjusting outliers to a reasonable range based on domain knowledge or statistical methods; statistical thresholds include but are not limited to standard scores (Z-Score) and interquartile ranges (Interquartile Range, IQR); machine learning methods include but are not limited to algorithms such as isolation forests and local anomaly factors.

[0055] For example, for the fare feature, if a non-numeric abnormal value such as 700AC appears, its suffix letters are removed and the numerical value is retained.

[0056] The above-mentioned preset feature range refers to the numerical range used when standardizing or normalizing specific feature data during the data preprocessing stage. The purpose of the preset feature range is to ensure that all data are compared and analyzed at the same scale to avoid model training bias caused by dimensional differences. Common preset feature ranges include between 0 and 1 or between -1 and 1. By applying linear transformations such as Min-Max Scaling or MaxAbs Scaling, the data is adjusted to the preset range, thereby improving the training efficiency and prediction accuracy of the model.

[0057] The above-mentioned normalization process refers to scaling the numerical features in the historical behavior data after data cleaning to a specific range to eliminate the differences in dimensions and numerical ranges between different features, so that all features have the same scale and influence in model training.

[0058] Exemplarily, the data eigenvalues ​​may be scaled to the range of (0, 1) by using Formula 1 to eliminate the impact of differences in the ranges of different eigenvalues.

[0059]

[0060] Where: X is the original eigenvalue, X′ is the standardized eigenvalue, μ is the mean of the eigenvalue, and σ is the standard deviation of the eigenvalue.

[0061] The above-mentioned feature extraction refers to the process of generating a set of feature vectors that can more accurately reflect the passenger's behavior patterns and preferences from the historical behavior data that has been cleaned and normalized. Feature extraction includes but is not limited to the following steps: numerical feature processing, categorical feature processing, and time feature processing.

[0062] For example, for numerical features, such as the ticket price and flight mileage purchased by passengers, mathematical transformation or creation of new derived features, such as calculating the ratio of the ticket price to the average ticket price, can be used to obtain the relationship and potential patterns between the numerical values; for categorical features, such as frequent flyer levels (silver card, gold card, platinum card, etc.), one-hot encoding is performed, that is, the categorical variables are converted into binary vectors, and each category corresponds to an element in the vector, to ensure that the relationship between the categories is not misled by the size of the values; for time features, time-related features are extracted, such as the ticket purchase date, the month of the flight date, the hour of the day, etc., to help the model understand the behavior patterns of passengers in a specific time period, such as whether there is seasonal ticket purchasing behavior, whether there is a preference for choosing additional services at a specific time point, etc.

[0063] The above correlation analysis parameters include but are not limited to Pearson correlation coefficient, Spearman correlation coefficient, Kendall rank correlation coefficient, etc., which are used to evaluate whether there is a statistical correlation between the feature and the target variable.

[0064] The above importance evaluation parameters are used to evaluate the relative importance of features in the prediction model. In machine learning, feature importance evaluation methods include but are not limited to feature importance based on tree models (such as random forests, gradient boosting trees, etc.), analysis based on model weights, analysis based on local explanations, analysis based on the contribution of features to model prediction results, etc.

[0065] For example, the feature extraction results are screened using correlation analysis parameters or importance assessment parameters to remove redundant or irrelevant features, and features with a high proportion of null values ​​and a high proportion of a certain feature value are removed. The feature screening results are then used to construct a data set, and the data set is divided into a training set, a validation set, and a test set according to a split ratio of 7:2:1 to facilitate model training, validation, and testing.

[0066] Based on the above optional embodiments, data quality can be improved while highlighting key feature information through missing value processing, outlier processing, normalization processing, feature extraction and feature screening.

[0067] Optionally, in step S13, the target prediction network model includes: an encoding module and an analysis module, and the target prediction network model is used to analyze and process the preprocessing result to obtain the user intention prediction result including:

[0068] Step S131, using an encoding module to convert the preprocessing result to obtain a first feature vector;

[0069] Step S132: Analyze and process the first feature vector using an analysis module to obtain a user intention prediction result.

[0070] The above conversion process refers to the process of converting the preprocessed data into a feature vector form that can be directly used by the machine learning model, which is usually called feature vectorization or embedding. In the conversion process, each row of data, that is, the feature set of each passenger, is converted into a fixed-length numerical vector.

[0071] For example, for categorical features, such as the frequent flyer card level of a passenger (silver card, gold card, platinum card), one-hot encoding can be used in the conversion process to convert each category into an independent binary feature. For example, the silver card is encoded as [1,0,0], the gold card is encoded as [0,1,0], and the platinum card is encoded as [0,0,1], so that the model can understand different levels of membership in numerical form. Further, the first feature vector obtained after the conversion process is input into the analysis module for analysis and processing to obtain the user intention prediction result.

[0072] Based on the above optional embodiment, converting the preprocessed data into the first eigenvector through the encoding module can enhance the interpretability and predictive ability of the model, while improving computing efficiency and reducing the risk of overfitting.

[0073] Optionally, in step S132, the analysis module includes: a linear layer and a normalized exponential function layer, and the first feature vector is analyzed and processed by the analysis module to obtain a user willingness prediction result including:

[0074] Step S1321, using a linear layer to perform a linear transformation on the first eigenvector to obtain a first transformation result;

[0075] Step S1322, using a normalized exponential function layer to perform probability calculation on the first transformation result to obtain a probability distribution result;

[0076] Step S1323, performing classification processing based on the probability distribution result to obtain a classification result;

[0077] Step S1324, determining the user intention prediction result based on the classification result.

[0078] The above linear transformation refers to inputting the feature vector into a linear layer for transformation, which maps the input feature vector to an output vector through the weight matrix (W) and the bias vector (b). The linear transformation can be performed using Formula 2.

[0079] y=Wx+b Formula 2

[0080] Where x is the input feature vector, W is the weight matrix, b is the bias vector, and y is the output vector.

[0081] The above probability calculation refers to converting the output vector obtained by linear transformation through a normalized exponential function layer (Softmax layer) to obtain the probability distribution of each category. The Softmax function can convert the numerical values ​​in the output vector into probability values, ensuring that the sum of the probabilities of all categories is 1, and the category corresponding to the maximum value in the output vector has the highest probability. The output vector y contains two elements, which represent the probability or score of the sample belonging to the positive class (purchase additional services) and the negative class (do not purchase additional services). The output vector can be converted into a probability distribution using Formula 3 to interpret the probabilities of the two elements as positive and negative classes.

[0082]

[0083] Among them, P(class=i) is the probability that the sample belongs to class i, y i is the value of the i-th element in the output vector.

[0084] The above classification process refers to classifying the sample into positive or negative classes according to the size relationship between the two elements of the output vector. If the value of the first element is greater than the value of the second element, it is predicted that the sample belongs to the positive class, that is, the user purchases the additional service; otherwise, it is predicted that it belongs to the negative class, that is, the user does not purchase the additional service.

[0085] For example, since there are a large number of negative classes in the sample data, such sparse data will affect the accuracy of cabin space prediction. Therefore, the samples with an output result of 1 (i.e., positive class) in the purchase intention prediction module are selected, and a certain proportion of negative classes are selected at the same time to reduce the sparse proportion of the samples and then input into the model, thereby enhancing the robustness of the model.

[0086] Based on the above optional embodiment, a linear layer is used to perform a linear transformation on the first eigenvector to obtain a first transformation result, and then a Softmax layer is used to perform probability calculation on the first transformation result to obtain a probability distribution result of purchasing additional services and not purchasing additional services, and then classification processing is performed based on the probability distribution result to obtain a classification result, and a user intention prediction result is determined based on the classification result, which can achieve accurate prediction of passengers' willingness to purchase additional services.

[0087] Optionally, in step S14, the user willingness prediction result is subjected to dimensionality reduction processing, and the dimensionality reduction result obtained includes:

[0088] Step S141, converting the user intention prediction result to obtain a second feature vector;

[0089] Step S142, performing a linear transformation on the second eigenvector to obtain a second transformation result;

[0090] Step S143, performing splicing processing on the second transformation result to obtain a feature splicing result;

[0091] Step S144, compressing the feature concatenation result to obtain a dimension reduction result.

[0092] For example, the user intention prediction result is converted into a feature vector [x1, x2, ..., x n ], where n represents the total number of features, Batch represents the batch size, batch = 1024. Use Formula 4 to perform linear transformation on the second eigenvector to obtain the second transformation result.

[0093] z i =W1x i +b1 Formula 4

[0094] Among them, i∈[1,n], n represents the total number of features; x i is the input feature vector, b1 is the bias vector, W1 is the weight matrix, z i is the intermediate representation after linear mapping,

[0095] The above-mentioned concatenation process refers to combining multiple features into a large feature vector. By concatenating features, we can make full use of various information and improve the model's expressiveness and prediction effect. The intermediate representation z after linear transformation can be expressed as i The input represents the concatenation layer.

[0096] z combined =[z1,z i ,…,z n ] Formula 5

[0097] Among them, n represents the total number of features, Z combined is an intermediate representation containing all features z1 to z n The new eigenvector of

[0098] The compression process is called dimensionality reduction. The goal of dimensionality reduction is to retain the important information of the original features while removing redundancy and noise, so that the feature vector after dimensionality reduction does not significantly reduce its expressive power, but significantly reduces its computational complexity. The feature vector z after concatenation can be expressed by formula 6: combined The shape is adjusted to z combined , which is input into the compression layer,

[0099] z′=W2z c ' ombined +b2 Formula 6

[0100] Among them, z′ is the eigenvector after dimensionality reduction, b2 is the bias vector, W2 is the weight matrix of the dimension reduction layer,

[0101] Based on the above optional embodiments, the user willingness prediction result is converted to obtain a second eigenvector, and the second eigenvector is linearly transformed to obtain a second transformation result, and then the second transformation result is spliced ​​to obtain a feature splicing result, and finally the feature splicing result is compressed to obtain a dimensionality reduction result. This can achieve efficient representation and dimensionality reduction of features, thereby effectively reducing the complexity of subsequent calculations, accelerating the model processing speed, retaining key information, enhancing the generalization ability of the model, and improving the recommendation accuracy.

[0102] Optionally, in step S15, generating target recommendation information based on the dimension reduction result includes:

[0103] Step S151, performing attention calculation based on the dimensionality reduction result to obtain an attention calculation result;

[0104] Step S152, performing residual normalization calculation according to the attention calculation result to obtain a first calculation result;

[0105] Step S153: input the first calculation result into the target deep neural network model for analysis and processing to generate target recommendation information.

[0106] The above attention calculation refers to the use of the attention mechanism to evaluate the importance of each part of the feature vector based on the dimensionality reduction result, so as to dynamically adjust the model's weight distribution of input features. In deep learning, given a query, the attention mechanism uses attention pooling to bias the selection towards sensory input. Sensory input, that is, values. Each value is paired with a key, the key is used to calculate the attention distribution, and the value is used to calculate the aggregate information. Specifically, the attention calculation assigns a weight to each feature vector element or each input sub-vector through an attention scoring function, indicating its contribution to the model prediction. The above weights are used to weighted sum the feature vector elements to obtain a weighted aggregate representation, that is, the attention calculation result.

[0107] For example, attention calculation can be performed by Formula 7:

[0108]

[0109] Among them, u is the final aggregate representation obtained, which integrates all vectors v i important information, focusing on the information of high-weight vectors; attn is the impact factor; q and k i The attention weight is obtained by mapping the two vectors into scalars through the attention scoring function and then through softmax, which can be calculated by formula 8:

[0110]

[0111] Among them, s is the attention scoring function, which can be expressed in the form of scaled dot product by formula 9:

[0112]

[0113] Wherein, D is the dimension of the feature vector, and in the embodiment of the present disclosure, D=256.

[0114] The above residual normalization calculation refers to the steps of performing residual connection and layer normalization on the feature vector after the attention calculation. Residual connection is the core idea of ​​the residual network. It improves the efficiency of information propagation, prevents gradient explosion and gradient disappearance, and speeds up model convergence by adding direct connections (Shortcut Connection) (also called residual connection) to the nonlinear convolutional layer. Layer normalization is a regularization technique used to accelerate model training and improve its stability. By calculating the mean and variance of the output of each layer of neurons, the output is normalized to make the output distribution of data in each layer more stable and reduce the problem of internal covariate shift. Residual normalization calculation can be performed using Formula 10 and Formula 11:

[0115] y1=z′+Dropout(u) Formula 10

[0116] y2=LayerNorm(y1) Formula 11

[0117] Among them, z′ is the feature vector after dimensionality reduction, u is the final aggregate representation obtained by attention calculation, and Dropout is a regularization technique that reduces the overfitting of the neural network by randomly discarding neurons during training.

[0118] Furthermore, the first calculation result obtained after processing in the above steps is input into the target deep neural network model for analysis and processing to generate target recommendation information.

[0119] Based on the above optional embodiment, a first calculation result is obtained through attention calculation and residual normalization calculation, and then the first calculation result is input into the target deep neural network model for analysis and processing to generate target recommendation information, which can improve the accuracy and efficiency of the recommendation system while enhancing the interpretability and robustness of the model.

[0120] Optionally, in step S153, the target deep neural network model includes: multiple convolutional layers, rectified linear unit layers, and fully connected layers, and the first calculation result is input into the target deep neural network model for analysis and processing, and generating target recommendation information includes:

[0121] Step S1531, using a first convolution layer to perform convolution processing on the first calculation result to obtain a first convolution processing result;

[0122] Step S1532, using a rectified linear unit layer to filter the first convolution processing result to obtain a target filtering result;

[0123] Step S1533, using the second convolution layer to perform convolution processing on the target screening result to obtain a second convolution processing result;

[0124] Step S1534, performing residual normalization processing according to the second convolution processing result to obtain a second calculation result;

[0125] Step S1535, using a fully connected layer to perform prediction processing on the second calculation result to obtain a target prediction result;

[0126] Step S1536, generating target recommendation information based on the target prediction result.

[0127] The above first convolution processing result can be calculated by formula 12:

[0128] y3=Dropout(RELU(Conv1(y2))) Formula 12

[0129] Among them, RELU is a commonly used activation function, which is widely used in deep learning models; Conv1 is the first convolutional layer, which is a one-dimensional convolutional layer with an input size of 256, a convolution kernel size of 1, and an output of 1024.

[0130] The above target screening results can be calculated by formula 13:

[0131] ReLU(x)=max(0,x) Formula 13

[0132] Among them, x is the input value, and ReLU(x) is the output value after activation, that is, the target screening result.

[0133] Furthermore, the target screening result is convolved by the second convolution layer to obtain the second convolution result, as shown in Formula 14:

[0134] y4=Dropout(Conv2(y3)) Formula 14

[0135] Among them, Conv2, the second convolutional layer, is a one-dimensional convolutional layer with an input size of 1024, a convolution kernel size of 1, and an output of 256.

[0136] Furthermore, the residual normalization process is performed on the second convolution processing result using formula 15 to obtain the second calculation result:

[0137] y5=LayerNorm(y2+y4) Formula 15

[0138] Finally, a fully connected layer is used to predict the type and level of additional services that the user may purchase, and the Softmax function is used to calculate the probability distribution of each category, as shown in Formula 16:

[0139] P=SoftMax(FullConnect(y5)) Formula 16

[0140] Among them, P is the probability distribution of each category, and the category with the highest probability is selected as the recommendation result; FullConnect is a fully connected layer with an input dimension of 256 and an output dimension of the total number of categories of additional services.

[0141] Exemplarily, the problem of gradient vanishing and gradient exploding in deep neural network training is solved by residual network, so that the network can be trained more efficiently and reach a deeper level. Specifically, two residual and normalization calculations are performed, and two one-dimensional convolutional layers are added between the above two calculations to enhance the representation ability of the deep neural network. First, after the first residual and normalization calculation, the first one-dimensional convolutional layer is used. The input size of this convolutional layer is 256, the convolution kernel size is 1, and the output size is 1024. Next, it is processed by the second one-dimensional convolutional layer with an input size of 1024, a convolution kernel size of 1, and an output size of 256. Then, a second residual and normalization calculation is performed. Finally, the results of the second residual and normalization calculation are input to the fully connected layer, and the probability distribution of additional services that the user may purchase is obtained through the softmax function.

[0142] Based on the above optional embodiments, through convolution processing, screening processing, and residual normalization processing, not only the gradient problem in the deep network is alleviated, but also the expression ability and prediction accuracy of the model are improved.

[0143] Figure 2is a schematic diagram of an information recommendation model according to one embodiment of the present disclosure, such as Figure 2 As shown in the figure, the model includes two modules, the purchase intention prediction module and the recommendation value and influence factor calculation module. The main function of the purchase intention prediction module is to predict the user's willingness to purchase additional services by analyzing the user's behavior data. The main function of the recommendation value prediction and influence factor calculation module is to output the influence factor and cabin recommendation through representation concatenation (integration), representation compression layer (dimensionality reduction), attention calculation, first residual normalization calculation, deep neural network representation and second residual normalization calculation, so as to predict the proportion of cabin recommendation and specific recommendation results.

[0144] Figure 3 is a structural block diagram of an information recommendation device according to one embodiment of the present disclosure, such as Figure 3 As shown, the device comprises:

[0145] The acquisition module 301 is used to acquire the historical behavior data of the target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user;

[0146] A preprocessing module 302 is used to preprocess the historical behavior data to obtain a preprocessing result;

[0147] The first processing module 303 is used to analyze and process the preprocessing result using a target prediction network model to obtain a user intention prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data;

[0148] The second processing module 304 is used to perform dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result;

[0149] The generating module 305 is used to generate target recommendation information based on the dimension reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0150] Optionally, the preprocessing module 302 is also used to: perform data cleaning processing on the historical behavior data to obtain data cleaning results, wherein the data cleaning processing includes at least one of the following: missing value processing and outlier processing; normalizing the data cleaning results according to a preset feature range to obtain a normalized result; performing feature extraction based on the normalized result to obtain a feature extraction result; filtering the feature extraction results using preset filtering parameters to obtain a feature filtering result, wherein the preset filtering parameters include at least one of the following: correlation analysis parameters or importance assessment parameters; segmenting the feature filtering results to obtain a preprocessing result.

[0151] Optionally, the first processing module 303 is further used to: use the encoding module to convert the preprocessing result to obtain a first feature vector; use the analysis module to analyze the first feature vector to obtain a user willingness prediction result.

[0152] Optionally, the first processing module 303 is also used to: use a linear layer to perform a linear transformation on the first eigenvector to obtain a first transformation result; use a normalized exponential function layer to perform probability calculation on the first transformation result to obtain a probability distribution result; perform classification processing based on the probability distribution result to obtain a classification result; and determine a user intention prediction result based on the classification result.

[0153] Optionally, the second processing module 304 is also used to: convert the user intention prediction result to obtain a second eigenvector; perform linear transformation on the second eigenvector to obtain a second transformation result; perform splicing processing on the second transformation result to obtain a feature splicing result; perform compression processing on the feature splicing result to obtain a dimensionality reduction result.

[0154] Optionally, the generation module 305 is also used to: perform attention calculation based on the dimensionality reduction result to obtain an attention calculation result; perform residual normalization calculation based on the attention calculation result to obtain a first calculation result; input the first calculation result into the target deep neural network model for analysis and processing to generate target recommendation information.

[0155] Optionally, the generation module 305 is also used to: use the first convolution layer to convolve the first calculation result to obtain a first convolution processing result; use the rectified linear unit layer to filter the first convolution processing result to obtain a target filtering result; use the second convolution layer to convolve the target filtering result to obtain a second convolution processing result; perform residual normalization according to the second convolution processing result to obtain a second calculation result; use the fully connected layer to predict the second calculation result to obtain a target prediction result; and generate target recommendation information based on the target prediction result.

[0156] According to another aspect of an embodiment of the present disclosure, an electronic device is further provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the information recommendation method in the embodiment of the present disclosure.

[0157] Optionally, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0158] S1, obtaining historical behavior data of a target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user;

[0159] S2, preprocessing the historical behavior data to obtain preprocessing results;

[0160] S3, analyzing and processing the preprocessing results using a target prediction network model to obtain a user intention prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data;

[0161] S4, performing dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result;

[0162] S5, generating target recommendation information based on the dimension reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0163] According to another aspect of an embodiment of the present disclosure, a computer-readable storage medium is further provided. The computer-readable storage medium includes a stored executable program, wherein when the executable program runs, the device where the storage medium is located is controlled to execute the information recommendation method in the embodiment of the present disclosure.

[0164] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0165] S1, obtaining historical behavior data of a target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user;

[0166] S2, preprocessing the historical behavior data to obtain preprocessing results;

[0167] S3, analyzing and processing the preprocessing results using a target prediction network model to obtain a user intention prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data;

[0168] S4, performing dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result;

[0169] S5, generating target recommendation information based on the dimension reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0170] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store computer programs.

[0171] According to another aspect of an embodiment of the present disclosure, a computer program product is further provided. The computer program product includes computer instructions. When the computer instructions are executed by a processor, the information recommendation method in the embodiment of the present disclosure is implemented.

[0172] Optionally, in this embodiment, the computer program product may be configured as a computer program for executing the following steps:

[0173] S1, obtaining historical behavior data of a target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user;

[0174] S2, preprocessing the historical behavior data to obtain preprocessing results;

[0175] S3, analyzing and processing the preprocessing results using a target prediction network model to obtain a user intention prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data;

[0176] S4, performing dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result;

[0177] S5, generating target recommendation information based on the dimension reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

[0178] The serial numbers of the above-mentioned embodiments of the present disclosure are only for description and do not represent the advantages or disadvantages of the embodiments.

[0179] In the above embodiments of the present disclosure, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0180] In the several embodiments provided in the present disclosure, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0181] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0182] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0183] 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 technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or optical disk and other media that can store program codes.

[0184] The above is only a preferred embodiment of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present disclosure. These improvements and modifications should also be regarded as the scope of protection of the present disclosure.

Claims

1. An information recommendation method, characterized in that: include: Acquire historical behavior data of the target user, wherein the historical behavior data includes: user attribute information and historical transaction information associated with the target user; Preprocessing the historical behavior data to obtain a preprocessing result; Analyzing and processing the preprocessing results using a target prediction network model to obtain a user intention prediction result, wherein the target prediction network model is obtained by machine learning training using multiple groups of sample behavior data; Performing dimensionality reduction processing on the user willingness prediction result to obtain a dimensionality reduction result; Target recommendation information is generated based on the dimension reduction result, wherein the target recommendation information is used to represent the cabin recommendation content and feature influencing parameters corresponding to the target user.

2. The information recommendation method according to claim 1, characterized in that: Preprocessing the historical behavior data to obtain the preprocessing results includes: Performing data cleaning processing on the historical behavior data to obtain a data cleaning result, wherein the data cleaning processing includes at least one of the following: missing value processing and outlier value processing; Normalizing the data cleaning result according to a preset feature range to obtain a normalized result; Performing feature extraction based on the normalization result to obtain a feature extraction result; The feature extraction result is screened using preset screening parameters to obtain a feature screening result, wherein the preset screening parameters include at least one of the following: a correlation analysis parameter or an importance assessment parameter; The feature screening result is segmented to obtain the preprocessing result.

3. The information recommendation method according to claim 1, characterized in that: The target prediction network model includes: an encoding module and an analysis module. The target prediction network model is used to analyze and process the preprocessing result to obtain the user willingness prediction result, which includes: Using the encoding module to convert the preprocessing result to obtain a first feature vector; The first feature vector is analyzed and processed by using the analysis module to obtain the user intention prediction result.

4. The information recommendation method according to claim 3, characterized in that: The analysis module includes: a linear layer and a normalized exponential function layer. The analysis module is used to analyze and process the first feature vector to obtain the user willingness prediction result including: Using the linear layer to perform a linear transformation on the first feature vector to obtain a first transformation result; Using the normalized exponential function layer to perform probability calculation on the first transformation result to obtain a probability distribution result; Perform classification processing based on the probability distribution result to obtain a classification result; The user intention prediction result is determined according to the classification result.

5. The information recommendation method according to claim 1, characterized in that: The user willingness prediction result is subjected to dimensionality reduction processing, and the dimensionality reduction result obtained includes: Converting the user intention prediction result to obtain a second feature vector; Performing a linear transformation on the second eigenvector to obtain a second transformation result; Performing splicing processing on the second transformation results to obtain a feature splicing result; The feature splicing result is compressed to obtain the dimension reduction result.

6. The information recommendation method according to claim 1, characterized in that: Generating the target recommendation information based on the dimension reduction result includes: Performing attention calculation based on the dimensionality reduction result to obtain an attention calculation result; Perform residual normalization calculation according to the attention calculation result to obtain a first calculation result; The first calculation result is input into the target deep neural network model for analysis and processing to generate the target recommendation information.

7. The information recommendation method according to claim 6, characterized in that: The target deep neural network model includes: multiple convolutional layers, rectified linear unit layers, and fully connected layers. The first calculation result is input into the target deep neural network model for analysis and processing. Generating the target recommendation information includes: Using a first convolution layer to perform convolution processing on the first calculation result to obtain a first convolution processing result; Using the rectified linear unit layer to perform screening processing on the first convolution processing result to obtain a target screening result; Using a second convolutional layer to perform convolution processing on the target screening result to obtain a second convolution processing result; Performing residual normalization processing according to the second convolution processing result to obtain a second calculation result; Using the fully connected layer to perform prediction processing on the second calculation result to obtain a target prediction result; The target recommendation information is generated based on the target prediction result.

8. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the information recommendation method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the storage medium is located is controlled to execute the information recommendation method according to any one of claims 1 to 7.

10. A computer program product, characterized in that The computer program product comprises computer instructions, which, when executed by a processor, implement the information recommendation method as claimed in any one of claims 1 to 7.

Images