Information recommendation method, device and computer equipment

Abstract

The present application relates to an information recommendation method, device, computer equipment, storage medium and computer program product, which are applied to the field of artificial intelligence technology. The method comprises: obtaining information features of the information to be recommended, performing domain classification on the information features, and obtaining feature domains corresponding to the information features; determining the mutual information value between feature domains corresponding to information feature combinations of different feature domains according to the feature domains corresponding to the information features; the information feature combination comprises two information features; inputting the information features and the mutual information value between feature domains corresponding to the information feature combinations of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended; and determining the recommendation result of the information to be recommended according to the recommendation degree. The use of this method is conducive to improving the accuracy of information recommendation.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an information recommendation method, apparatus, computer equipment, storage medium and computer program product. Background Art

[0002] With the development of Internet technology, there is more and more information on the Internet. The recommendation system will recommend information to users based on user requests.

[0003] In traditional technology, when recommending information, the user's preference for information is generally predicted based on the attribute characteristics of the information, and the information is recommended to the user when the user has a clear preference for the information; however, the user's preference for information is affected by many factors, and only considering the attribute characteristics of the information will result in a low accuracy rate in information recommendation. Summary of the invention

[0004] Based on this, it is necessary to provide an information recommendation method, device, computer equipment, computer-readable storage medium and computer program product that can improve the accuracy of information recommendation in response to the above technical problems.

[0005] In a first aspect, the present application provides an information recommendation method. The method comprises:

[0006] Acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features;

[0007] According to the feature domains corresponding to the information features, determining mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains; the information feature combination includes two information features;

[0008] Inputting the mutual information value between the information feature and the feature domains corresponding to the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended;

[0009] According to the recommendation degree, a recommendation result of the information to be recommended is determined.

[0010] In one embodiment, determining, according to the feature domains corresponding to the information features, the mutual information values ​​between the feature domains corresponding to the information feature combinations of different feature domains comprises:

[0011] Determining information feature combinations of different feature domains according to feature domains corresponding to the information features;

[0012] Counting the individual occurrence frequency of each feature domain corresponding to the information feature combination of the different feature domains, and the common occurrence frequency of the corresponding feature domains;

[0013] According to the individual occurrence frequency of each feature domain corresponding to the information feature combination of the different feature domains and the common occurrence frequency of the corresponding feature domains, the mutual information value between the feature domains corresponding to the information feature combination of the different feature domains is obtained.

[0014] In one embodiment, obtaining the mutual information value between the feature domains corresponding to the information feature combination of the different feature domains according to the individual occurrence frequency of each feature domain corresponding to the information feature combination of the different feature domains and the common occurrence frequency of the corresponding feature domains includes:

[0015] Determine the product of the individual occurrence frequencies of each feature domain corresponding to the information feature combination of the different feature domains;

[0016] According to the ratio between the common occurrence frequency of the feature domains corresponding to the information feature combinations of the different feature domains and the product, query the corresponding relationship between the ratio and the feature domain association degree to obtain the association degree between the feature domains corresponding to the information feature combinations of the different feature domains;

[0017] The correlation degree between the feature domains corresponding to the information feature combinations of the different feature domains is determined as the mutual information value between the feature domains corresponding to the information feature combinations of the different feature domains.

[0018] In one embodiment, inputting the mutual information value between the information feature and the feature domains corresponding to the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended includes:

[0019] Obtaining an information feature vector of the information feature through a first processing model in the pre-trained information recommendation model;

[0020] Processing the N-order information feature combination according to the information feature vector to obtain a first output result; N is greater than or equal to 3;

[0021] Processing the information feature, the information feature combination, the information feature vector and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains through the second processing model in the pre-trained information recommendation model to obtain a second output result;

[0022] A weighted sum is performed on the first output result and the second output result to obtain a recommendation degree of the information to be recommended.

[0023] In one embodiment, the second processing model in the pre-trained information recommendation model processes the information feature, the information feature combination, the information feature vector and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains to obtain a second output result, including:

[0024] Determine, by means of a second processing model in the pre-trained information recommendation model, a weight corresponding to the information feature, and determine, as a weight corresponding to the information feature combination, a product of an inner product of an information feature vector corresponding to an information feature in the information feature combination and a mutual information value between feature domains corresponding to the information feature combination;

[0025] Obtaining a first product result between the information feature and the weight corresponding to the information feature, and a second product result between the information feature combination and the weight corresponding to the information feature combination;

[0026] The first multiplication result and the second multiplication result are added to obtain the second output result.

[0027] In one embodiment, the pre-trained information recommendation model is trained in the following manner:

[0028] Acquire training data; the training data includes sample information features of sample information, mutual information values ​​between feature domains corresponding to combinations of sample information features of different feature domains, and actual recommendation degrees of the sample information;

[0029] According to the training data, jointly training the first processing model and the second processing model in the information recommendation model to be trained until a training end condition is reached;

[0030] The information recommendation model that meets the training end condition is determined as the pre-trained information recommendation model.

[0031] In one embodiment, the step of jointly training the first processing model and the second processing model in the information recommendation model to be trained according to the training data until a training end condition is reached includes:

[0032] Inputting the mutual information value between the feature domains corresponding to the sample information feature and the combination of the sample information features of the different feature domains into the information recommendation model to be trained to obtain the predicted recommendation degree of the sample information;

[0033] Calculating a loss value according to a difference between the predicted recommendation degree and the actual recommendation degree;

[0034] When the loss value is greater than or equal to a preset threshold, the model parameters of the first processing model and the second processing model in the information recommendation model to be trained are adjusted according to the loss value to obtain an adjusted information recommendation model, and the adjusted information recommendation model is repeatedly trained until the loss value obtained according to the trained information recommendation model is less than the preset threshold.

[0035] In one embodiment, determining the recommendation result of the information to be recommended according to the recommendation degree includes:

[0036] Filtering target information whose recommendation degree is greater than a preset recommendation degree from the information to be recommended;

[0037] The target information is recommended to the corresponding user.

[0038] In a second aspect, the present application also provides an information recommendation device. The device comprises:

[0039] An acquisition module, used to acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features;

[0040] A statistical module, for determining, according to the feature domain corresponding to the information feature, mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains; the information feature combination includes two information features;

[0041] A prediction module, used for inputting the mutual information value between the feature domains corresponding to the information feature and the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended;

[0042] The determination module is used to determine the recommendation result of the information to be recommended according to the recommendation degree.

[0043] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0044] Acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features;

[0045] According to the feature domain corresponding to the information feature, determining the mutual information value between the feature domains corresponding to the information feature combination of different feature domains; the information feature combination includes two information features;

[0046] Inputting the mutual information value between the information feature and the feature domains corresponding to the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended;

[0047] According to the recommendation degree, a recommendation result of the information to be recommended is determined.

[0048] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0049] Acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features;

[0050] According to the feature domain corresponding to the information feature, determining the mutual information value between the feature domains corresponding to the information feature combination of different feature domains; the information feature combination includes two information features;

[0051] Inputting the mutual information value between the information feature and the feature domains corresponding to the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended;

[0052] According to the recommendation degree, a recommendation result of the information to be recommended is determined.

[0053] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0054] Acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features;

[0055] According to the feature domain corresponding to the information feature, determining the mutual information value between the feature domains corresponding to the information feature combination of different feature domains; the information feature combination includes two information features;

[0056] Inputting the mutual information value between the information feature and the feature domains corresponding to the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended;

[0057] According to the recommendation degree, a recommendation result of the information to be recommended is determined.

[0058] The above-mentioned information recommendation method, device, computer equipment, storage medium and computer program product obtain information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features; then, according to the feature domains corresponding to the information features, determine the mutual information value between feature domains corresponding to information feature combinations of different feature domains; finally, input the information features and the mutual information value between feature domains corresponding to information feature combinations of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended, and determine the recommendation result of the information to be recommended according to the recommendation degree. In this way, when recommending information, the information features of the information to be recommended and the mutual information value between feature domains corresponding to information feature combinations of different feature domains are comprehensively considered, which is conducive to improving the accuracy of determining the recommendation degree of the information to be recommended, making the recommendation result obtained based on the recommendation degree more accurate, thereby improving the accuracy of information recommendation. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flowchart of an information recommendation method in one embodiment;

[0060] Figure 2 A schematic flow chart of the steps of obtaining mutual information values ​​of information feature combinations of different feature domains in one embodiment;

[0061] Figure 3 A schematic flow chart of an information recommendation method in another embodiment;

[0062] Figure 4 is a flowchart of an information recommendation method in yet another embodiment;

[0063] Figure 5 is a structural block diagram of an information recommendation device in one embodiment;

[0064] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0065] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0066] In one embodiment, Figure 1 As shown, a method for migrating an instruction library is provided. This embodiment uses the method applied to a server as an example for illustration. It can be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0067] Step S101, obtaining information features of the information to be recommended, performing domain classification on the information features, and obtaining feature domains corresponding to the information features.

[0068] Among them, the information to be recommended refers to candidate information that needs to be recommended, such as videos, live broadcasts, pictures, products, news information, etc.

[0069] Among them, information characteristics include basic attribute characteristics of the information to be recommended and operational behavior characteristics of the recommended information; operational behavior characteristics include but are not limited to browsing, collecting, liking, commenting and forwarding.

[0070] Among them, the feature domain corresponding to the information feature refers to the category to which the information feature belongs, such as gender, age, occupation, interest, etc.

[0071] Specifically, the server obtains information features of the information to be recommended from the database, performs domain classification on the information features according to the domain classification instruction, and obtains the domain classification corresponding to the information features.

[0072] For example, by introducing the concept of domain, features of the same nature can be classified into the same domain. For example, the text feature "occupation" needs to be one-hot encoded first, and 100 new features are generated after encoding. Then these 100 new features can all be classified into the domain of "occupation".

[0073] Furthermore, the server may also query the correspondence between the preset information features and the feature domain according to the information features of the information to be recommended, and obtain the feature domain of the information features of the information to be recommended.

[0074] Step S102, determining mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains according to feature domains corresponding to information features; the information feature combination includes two information features.

[0075] Among them, the information feature combination of different feature domains refers to a combination of information features of two different feature domains; the feature domain corresponding to the information feature combination refers to the feature domain corresponding to each information feature in the information feature combination.

[0076] Among them, the mutual information value between feature domains corresponding to the information feature combination of different feature domains is used to measure the combination weight corresponding to the information feature combination of different feature domains, specifically used to measure the degree of correlation between the feature domains corresponding to the information features in the information feature combination.

[0077] Specifically, the server combines information features of different feature domains according to feature domains corresponding to the information features to obtain information feature combinations of different feature domains; and uses a mutual information algorithm to count mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains.

[0078] Step S103, input the information feature and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains into the pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended.

[0079] The pre-trained information recommendation model refers to a network model used to predict the recommendation degree of information, including an improved FM (Factorization Machine) model and a deep learning model based on mutual information values.

[0080] It should be noted that the improved FM model based on mutual information value refers to a model obtained by improving the FM model. The first improvement refers to the introduction of the mutual information algorithm in the FM model to calculate the cross-information weights of adjacent features, replacing the original inner product algorithm, thereby reducing the computational cost of high-dimensional feature crosses; the second improvement refers to the introduction of the domain classification concept in the domain factorization (FFM) model in the FM model, thereby improving the fitting ability of the model.

[0081] The recommendation degree of the information to be recommended refers to the recommendation probability of the information to be recommended; the higher the recommendation degree, the recommendation probability of the information to be recommended, and further indicates that the user prefers the information to be recommended.

[0082] Specifically, the server inputs the mutual information values ​​between the information features and the feature domains corresponding to the information feature combinations of different feature domains into a pre-trained information recommendation model, and performs a series of processing on the mutual information values ​​between the information features and the feature domains corresponding to the information feature combinations of different feature domains through the pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended.

[0083] Step S104: determining the recommendation result of the information to be recommended according to the recommendation degree.

[0084] Specifically, when the recommendation degree is greater than a preset recommendation degree, the server recommends the information to be recommended to the corresponding user.

[0085] In the above information recommendation method, by obtaining the information features of the information to be recommended, the information features are domain classified to obtain the feature domains corresponding to the information features; then, according to the feature domains corresponding to the information features, the mutual information values ​​between the feature domains corresponding to the information feature combinations of different feature domains are determined; the information feature combinations include two information features; finally, the information features and the mutual information values ​​between the feature domains corresponding to the information feature combinations of different feature domains are input into the pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended, and according to the recommendation degree, the recommendation result of the information to be recommended is determined. In this way, when recommending information, the information features of the information to be recommended and the mutual information values ​​between the feature domains corresponding to the information feature combinations of different feature domains are comprehensively considered, which is conducive to improving the accuracy of determining the recommendation degree of the information to be recommended, making the recommendation result obtained based on the recommendation degree more accurate, thereby improving the accuracy of information recommendation.

[0086] In one embodiment, Figure 3 As shown, the above step S102, according to the feature domains corresponding to the information features, determines the mutual information values ​​between the feature domains corresponding to the information feature combinations of different feature domains, and specifically includes the following steps:

[0087] Step S201, determining information feature combinations of different feature domains according to feature domains corresponding to information features.

[0088] Step S202 , counting the individual occurrence frequency of each feature domain corresponding to the information feature combination of different feature domains, and the common occurrence frequency of the corresponding feature domains.

[0089] The individual occurrence frequency of each feature domain corresponding to the information feature combination of different feature domains is used to measure the individual occurrence frequency of the feature domain corresponding to each information feature in the information feature combination of different feature domains.

[0090] The common occurrence frequency of feature domains corresponding to information feature combinations of different feature domains is used to measure the number of common occurrences between feature domains corresponding to each information feature in the information feature combinations of different feature domains.

[0091] Step S203, obtaining mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains according to the individual occurrence frequency of each feature domain corresponding to information feature combinations of different feature domains and the common occurrence frequency of corresponding feature domains.

[0092] Specifically, the server combines information features of different feature domains according to feature domains corresponding to information features to obtain information feature combinations of different feature domains; counts the individual occurrence frequency of feature domains corresponding to each information feature in the information feature combinations of different feature domains, and the common occurrence frequency between feature domains corresponding to each information feature, to obtain the individual occurrence frequency of each feature domain corresponding to the information feature combinations of different feature domains, and the common occurrence frequency of the corresponding feature domains; obtains the product of the individual occurrence frequency of each feature domain corresponding to the information feature combinations of different feature domains, and obtains the ratio of the common occurrence frequency of feature domains corresponding to the information feature combinations of different feature domains to the product, performs logarithm operation on the ratio, and obtains the mutual information value between the feature domains corresponding to the information feature combinations of different feature domains.

[0093] Furthermore, based on the individual occurrence frequency of each feature domain corresponding to the information feature combination of different feature domains and the common occurrence frequency of the corresponding feature domains, the mutual information value between the feature domains corresponding to the information feature combination of different feature domains is obtained, specifically including: determining the product between the individual occurrence frequency of each feature domain corresponding to the information feature combination of different feature domains; based on the ratio between the common occurrence frequency of the feature domains corresponding to the information feature combination of different feature domains and the product, querying the correspondence between the ratio and the feature domain correlation degree, and obtaining the correlation degree between the feature domains corresponding to the information feature combination of different feature domains; determining the correlation degree between the feature domains corresponding to the information feature combination of different feature domains as the mutual information value between the feature domains corresponding to the information feature combination of different feature domains.

[0094] For example, mutual information is an important information measurement method in information theory. It can be regarded as the amount of information contained in a random variable about another random variable, which represents the degree of association between two variables X and Y. Its formula can be simplified as:

[0095]

[0096] Among them, I(x1,x2) represents the mutual information value between feature x1 and feature x2; p(x1,x2) is the frequency of co-occurrence of feature x1 and feature x2; p(x1) and p(x2) are the frequencies of separate occurrences of feature x1 and feature x2 respectively; the larger I(x1,x2) is, the stronger the correlation between feature x1 and feature x2 is. In the scenario of recommendation system, many features are correlated. For example, the mutual information value of the recent number of views and whether the same product is collected will be very high, which has a strong correlation and needs to be given a higher weight in the model. The concept of domain is used in this application, so the mutual information value between domain classifications needs to be calculated:

[0097]

[0098] Among them, I(f1,f2) represents the mutual information value between domain f1 and domain f2; p(f1,f2) is the frequency of co-occurrence of domain f1 and domain f2; p(f1) and p(f2) are the frequencies of separate occurrence of domain f1 and domain f2 respectively; the larger I(f1,f2) is, the stronger the correlation between domain f1 and domain f2 is.

[0099] In this embodiment, by calculating the mutual information value between the feature domains corresponding to the information feature combination of different feature domains as the mutual information value of the information feature combination of different feature domains, it is beneficial to introduce the mutual information value of the information feature combination of different feature domains when performing recommendation degree prediction in the subsequent time, so that the correlation degree between the information features in the information feature combination of different feature domains can be comprehensively considered, so that the obtained recommendation degree is more accurate.

[0100] In one embodiment, the above step S103, inputting the mutual information value between the feature domains corresponding to the information feature and the information feature combination of different feature domains into the pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended, specifically includes: obtaining the information feature vector of the information feature through the first processing model in the pre-trained information recommendation model; processing the N-order information feature combination according to the information feature vector to obtain a first output result; N is greater than or equal to 3; processing the information feature, the information feature combination, the information feature vector and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains through the second processing model in the pre-trained information recommendation model to obtain a second output result; and performing weighted summation on the first output result and the second output result to obtain the recommendation degree of the information to be recommended.

[0101] Among them, the first processing model refers to a deep learning model, and the second processing model refers to an improved FM model based on mutual information value.

[0102] The information feature refers to the original sparse feature, and the information feature vector of the information feature refers to the dense feature obtained by compressing and embedding the information feature through the dense feature layer (such as embedding).

[0103] Among them, the N-order information feature combination refers to a combination of high-order information features, specifically a combination of three or more information features. A single information feature refers to a first-order information feature; an information feature combination refers to a combination of second-order information features; both a single information feature and the second-order information features in the information feature combination refer to low-order information features. It should be noted that the low-order information features are crossed using the improved FM model, and the high-order information features are crossed by introducing the deep learning model.

[0104] Specifically, the server inputs the information feature and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains into a pre-trained information recommendation model, converts the information feature through the dense feature layer in the first processing model in the pre-trained information recommendation model to obtain the information feature vector of the information feature; obtains the N-order information feature combination, and processes the information feature vector of the information feature in the N-order information feature combination according to the information feature vector of the information feature to obtain a first output result; processes the information feature, the information feature vector, the information feature combination of different feature domains, and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains through the second processing model in the pre-trained information recommendation model to obtain a second output result; obtains a first product of the first output result and the corresponding weight, and a second product of the second output result and the corresponding weight; adds and normalizes the first product and the second product to obtain the recommendation degree of the information to be recommended.

[0105] Furthermore, the information features, information feature combinations, information feature vectors, and mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains are processed through the second processing model in the pre-trained information recommendation model to obtain a second output result, specifically including: determining the weights corresponding to the information features, and determining the product of the inner product of the information feature vectors corresponding to the information features in the information feature combination and the mutual information values ​​between the feature domains corresponding to the information feature combination as the weights corresponding to the information feature combination through the second processing model in the pre-trained information recommendation model; obtaining a first product result between the information features and the weights corresponding to the information features, and a second product result between the information feature combination and the weights corresponding to the information feature combination; and adding the first product result and the second product result to obtain a second output result.

[0106] For example, by observing a large number of samples, we can find that after some features are associated, the correlation between them and the label will increase. Compared with a single feature, these combined features often have a stronger impact on model training. The FM model takes advantage of this feature and improves the performance of the model by introducing combined features in the linear model. It can be expressed as,

[0107]

[0108] Among them, n represents the number of features of the sample, x i is the value of the feature of the i-th dimension, v iis the vector of the i-th dimension feature, and <·,·> represents the vector dot product. For the FFM model, assuming that the n features of the sample belong to f domains, then FFM has nf latent vectors, while in the FM model, there is only one latent vector for each dimension of the feature. Therefore, due to the increase in latent vectors, the computational complexity of the FFM model is much higher than that of the FM model. Therefore, based on FM and combining the characteristics of FFM, this scheme introduces the mutual information value of domain features, which improves the model effect while reducing the computational overhead, which can be expressed as:

[0109]

[0110] Among them, I(f i ,f j ) is the domain f to which the i-th dimension feature belongs i The domain f to which the j-th dimension feature belongs j The mutual information value of .

[0111] The deep learning model is a feedforward network model, which uses low-dimensional dense vectors as features and uses ReLu as the activation function.

[0112] a (l+1) =f(W (l) a l +b l );

[0113] Among them, a (l) 、b (l) , W (l) , f represent the input, bias term, parameter term and activation function of layer l respectively.

[0114] The weighted summation method is used to add the two features to obtain the recommendation degree:

[0115]

[0116] in, Output results for the Wide section, is the parameter output from the Deep part to the last layer of activation function output. The result of the last layer of activation function output on the Deep side is a (lf) ,b is the global bias term. It should be noted that the wide part refers to the improved FM model, and the deep part refers to the deep learning model.

[0117] In this embodiment, the mutual information values ​​of information features and information feature combinations of different feature domains are input into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended. Comprehensive consideration of the mutual information values ​​of information features and information feature combinations of different feature domains and the pre-trained information recommendation model are used, which is beneficial to improving the accuracy of determining the recommendation degree and further improving the accuracy of information recommendation.

[0118] In one embodiment, a pre-trained information recommendation model is trained in the following manner: obtaining training data; the training data includes sample information features of sample information, mutual information values ​​between feature domains corresponding to combinations of sample information features in different feature domains, and actual recommendation degrees of sample information; based on the training data, jointly training the first processing model and the second processing model in the information recommendation model to be trained until a training end condition is met; and determining the information recommendation model that meets the training end condition as the pre-trained information recommendation model.

[0119] The training end conditions include, but are not limited to, reaching the training times and the loss function reaching convergence.

[0120] Specifically, the server obtains sample information features of the sample information and an actual recommendation degree of the sample information; performs domain classification on the sample information features to obtain feature domains corresponding to the sample information features; combines sample information features of different feature domains according to the feature domains corresponding to the sample information features to obtain sample information feature combinations of different feature domains; uses a mutual information algorithm to count mutual information values ​​between feature domains corresponding to sample information feature combinations of different feature domains; uses the sample information features of the sample information, the mutual information values ​​between feature domains corresponding to sample information feature combinations of different feature domains, and the actual recommendation degree of the sample information as training data; and jointly trains a first processing model (such as a deep learning model) and a second processing model (such as an improved FM model based on mutual information values) in the information recommendation model to be trained according to the training data until a training end condition is reached; when the training end condition is reached, the information recommendation model that reaches the training end condition is determined as a pre-trained information recommendation model.

[0121] In this embodiment, the first processing model and the second processing model in the information recommendation model to be trained are jointly trained according to the training data, which is beneficial to improving the accuracy of the recommendation degree output by the trained information recommendation model, thereby improving the accuracy of information recommendation.

[0122] In one embodiment, the first processing model and the second processing model in the information recommendation model to be trained are jointly trained according to the training data until the training end condition is reached, specifically including: inputting the mutual information value between the feature domains corresponding to the sample information feature and the sample information feature combination of different feature domains into the information recommendation model to be trained to obtain the predicted recommendation degree of the sample information; calculating the loss value according to the difference between the predicted recommendation degree and the actual recommendation degree; when the loss value is greater than or equal to a preset threshold, adjusting the model parameters of the first processing model and the second processing model in the information recommendation model to be trained according to the loss value to obtain an adjusted information recommendation model, and repeatedly training the adjusted information recommendation model until the loss value obtained according to the trained information recommendation model is less than the preset threshold.

[0123] Specifically, the server inputs the mutual information values ​​between the feature domains corresponding to the sample information features and the sample information feature combinations of different feature domains into the information recommendation model to be trained, and converts the sample information features through the dense feature layer in the first processing model (such as the deep learning model) in the information recommendation model to be trained to obtain the sample information feature vector of the sample information features; obtains the N-order sample information feature combination, and processes the sample information feature vector of the sample information features in the N-order sample information feature combination according to the sample information feature vector of the sample information feature to obtain a first prediction output result; and processes the sample information features, the sample information feature vectors, the sample information feature combinations of different feature domains, and the feature vectors corresponding to the sample information feature combinations of different feature domains through the second processing model in the information recommendation model to be trained (such as the improved FM model based on the mutual information value). The mutual information value between the feature domains is processed to obtain a second prediction output result; a first product of the first prediction output result and the corresponding weight, and a second product of the second prediction output result and the corresponding weight are obtained; the first product and the second product are added and normalized to obtain the predicted recommendation degree of the sample information; according to the difference between the predicted recommendation degree and the actual recommendation degree, combined with the loss function, a loss value is statistically obtained; when the loss value is greater than or equal to a preset threshold, the model parameters of the first processing model and the second processing model in the information recommendation model to be trained are adjusted according to the loss value to obtain an adjusted information recommendation model, and the adjusted information recommendation model is repeatedly trained until the loss value obtained according to the trained information recommendation model is less than the preset threshold, then the training is stopped, and the trained information recommendation model is used as the pre-trained information recommendation model.

[0124] In this embodiment, when training the first processing model and the second processing model in the information recommendation model, the mutual information values ​​between the feature domains corresponding to the combination of sample information features and sample information features of different feature domains are introduced, and the high-order information features are combined through the first processing model, and the low-order information features are combined through the second processing model, so that the recommendation degree output by the trained information recommendation model is more accurate, further improving the accuracy of determining the recommendation degree.

[0125] In one embodiment, the above step S104, determining the recommendation result of the information to be recommended according to the recommendation degree, specifically includes: screening out target information with a recommendation degree greater than a preset recommendation degree from the information to be recommended; and recommending the target information to the corresponding user.

[0126] For example, the information to be recommended includes information A, information B, information C and information D, and the corresponding recommendation degrees are 0.9, 0.8, 0.7 and 0.4 respectively, and the preset recommendation degree is 0.5, which means that the recommendation degrees of information A, information B and information C are all greater than the preset recommendation degrees. In this case, information A, information B and information C are all used as target information, and the server recommends the target information [information A, information B, information C] to the corresponding user terminal according to the preset recommendation method, and displays information A, information B and information C through the user terminal to facilitate users to view their favorite information.

[0127] In this embodiment, target information with a recommendation degree greater than a preset recommendation degree is screened out from the information to be recommended, and the target information is recommended to the corresponding user, thereby achieving the purpose of accurate push and further improving the accuracy of information recommendation.

[0128] In one embodiment, Figure 3 As shown, another information recommendation method is provided, which is described by taking the method applied to a server as an example, and includes the following steps:

[0129] Step S301, obtaining training data; the training data includes sample information features of sample information, mutual information values ​​between feature domains corresponding to combinations of sample information features of different feature domains, and actual recommendation degrees of sample information.

[0130] Step S302: jointly train the first processing model and the second processing model in the information recommendation model to be trained according to the training data until a training end condition is met.

[0131] Step S303: determine the information recommendation model that has reached the training end condition as the pre-trained information recommendation model.

[0132] Step S304, obtaining information features of the information to be recommended, performing domain classification on the information features, and obtaining feature domains corresponding to the information features.

[0133] Step S305, determining information feature combinations of different feature domains according to feature domains corresponding to information features; and counting the individual occurrence frequency of each feature domain corresponding to the information feature combinations of different feature domains, and the common occurrence frequency of the corresponding feature domains.

[0134] Step S306, obtaining mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains according to the individual occurrence frequency of each feature domain corresponding to the information feature combinations of different feature domains and the common occurrence frequency of the corresponding feature domains.

[0135] Step S307, obtaining an information feature vector of the information feature through the first processing model in the pre-trained information recommendation model; processing the N-order information feature combination according to the information feature vector to obtain a first output result; N is greater than or equal to 3.

[0136] Step S308, processing the information features, information feature combinations, information feature vectors, and mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains through the second processing model in the pre-trained information recommendation model to obtain a second output result.

[0137] Step S309: performing weighted summation on the first output result and the second output result to obtain the recommendation degree of the information to be recommended.

[0138] Step S310, filtering out target information whose recommendation degree is greater than a preset recommendation degree from the information to be recommended; and recommending the target information to the corresponding user.

[0139] In the above-mentioned information recommendation method, when recommending information, the mutual information values ​​between the feature domains corresponding to the information features of the information to be recommended and the information feature combinations of different feature domains are comprehensively considered, which is beneficial to improving the accuracy of determining the recommendation degree of the information to be recommended, making the recommendation results based on the recommendation degree more accurate, thereby improving the accuracy of information recommendation.

[0140] In one embodiment, in order to more clearly illustrate the information recommendation method provided by the embodiment of the present application, the information recommendation method is specifically described below with a specific embodiment. Figure 4 As shown, the present application also provides an information recommendation method, which classifies features, crosses low-order features using an improved FM model, crosses high-order features using a deep learning model, and weights the two models together, so that the combined weight of the features can be captured more accurately while reducing certain computational costs. Specifically, it includes the following contents:

[0141] This application mainly improves the model of feature cross-processing used in existing recommendation systems from three aspects: first, domain classification of features is performed, and latent vectors are selected to calculate feature relationships; at the same time, the mutual information algorithm is used instead of the traditional vector inner product calculation method to reduce the calculation cost; secondly, a deep learning model is used for high-order features to improve the performance of the model in high-order cross-features; finally, FM and deep learning models are jointly trained to assist the recommendation system in making more accurate decisions.

[0142] refer to Figure 4 , obtain training data in S401; perform reasonable domain classification on the features of training data in S402; obtain the mutual information value of each domain in S403; perform embedding processing on the sparse classification features in S404 to obtain low-dimensional dense vectors; use the FM model based on mutual information value and the deep learning model to jointly train the data in S405; use the model to predict and store the prediction results in S406. For example, the trained model is:

[0143]

[0144] Among them, Y can be regarded as the preference of user u for product i. When Y = 1, there is a clear preference; when Y = 0, there is no clear preference. X can be regarded as the feature used for prediction, W wide and W deep are the weights based on the joint training of the improved FM model and the deep learning model, φ(X) and a,b are the parameter items obtained based on the training of the improved FM model and the deep learning model respectively.

[0145] The above information recommendation method reduces the computational cost of feature combination of low-order features by improving the mutual information value algorithm and the FM model. At the same time, it uses the DNN deep learning model for high-order features, and weightedly combines the two models for calculation, thereby increasing the weight of highly correlated features and greatly improving the fitting effect of the final model, thereby improving the effect of the recommendation system, effectively predicting user preferences, and making accurate recommendation decisions.

[0146] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0147] Based on the same inventive concept, the embodiment of the present application also provides an information recommendation device for implementing the above-mentioned information recommendation method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more information recommendation device embodiments provided below can refer to the limitations on the information recommendation method above, and will not be repeated here.

[0148] In one embodiment, Figure 5 As shown, an information recommendation device is provided, including: an acquisition module 510, a statistical module 520, a prediction module 530 and a determination module 540, wherein:

[0149] The acquisition module 510 is used to acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features.

[0150] The statistical module 520 is used to determine the mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains according to the feature domains corresponding to the information features; the information feature combination includes two information features.

[0151] The prediction module 530 is used to input the information feature and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains into the pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended.

[0152] The determination module 540 is used to determine the recommendation result of the information to be recommended according to the recommendation degree.

[0153] In one embodiment, the statistical module 520 is also used to determine the information feature combination of different feature domains based on the feature domains corresponding to the information features; to count the individual occurrence frequency of each feature domain corresponding to the information feature combination of different feature domains, and the common occurrence frequency of the corresponding feature domains; and to obtain the mutual information value between the feature domains corresponding to the information feature combination of different feature domains based on the individual occurrence frequency of each feature domain corresponding to the information feature combination of different feature domains and the common occurrence frequency of the corresponding feature domains.

[0154] In one embodiment, the statistical module 520 is also used to determine the product between the individual occurrence frequencies of each feature domain corresponding to the information feature combinations of different feature domains; based on the ratio between the common occurrence frequencies of the feature domains corresponding to the information feature combinations of different feature domains and the product, query the correspondence between the ratio and the feature domain correlation degree to obtain the correlation degree between the feature domains corresponding to the information feature combinations of different feature domains; determine the correlation degree between the feature domains corresponding to the information feature combinations of different feature domains as the mutual information value between the feature domains corresponding to the information feature combinations of different feature domains.

[0155] In one embodiment, the prediction module 530 is also used to obtain an information feature vector of an information feature through a first processing model in a pre-trained information recommendation model; process an N-order information feature combination according to the information feature vector to obtain a first output result; N is greater than or equal to 3; process information features, information feature combinations, information feature vectors, and mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains through a second processing model in the pre-trained information recommendation model to obtain a second output result; and perform weighted summation on the first output result and the second output result to obtain a recommendation degree of the information to be recommended.

[0156] In one embodiment, the prediction module 530 is also used to determine the weight corresponding to the information feature through the second processing model in the pre-trained information recommendation model, and determine the product of the inner product of the information feature vector corresponding to the information feature in the information feature combination and the mutual information value between the feature domains corresponding to the information feature combination as the weight corresponding to the information feature combination; obtain a first product result between the information feature and the weight corresponding to the information feature, and a second product result between the information feature combination and the weight corresponding to the information feature combination; add the first product result and the second product result to obtain a second output result.

[0157] In one embodiment, the information recommendation device also includes a training module for acquiring training data; the training data includes sample information features of the sample information, mutual information values ​​between feature domains corresponding to combinations of sample information features in different feature domains, and actual recommendation degrees of the sample information; based on the training data, the first processing model and the second processing model in the information recommendation model to be trained are jointly trained until a training end condition is met; the information recommendation model that meets the training end condition is determined as a pre-trained information recommendation model.

[0158] In one embodiment, the training module is also used to input the mutual information value between the feature domains corresponding to the sample information features and the combination of sample information features of different feature domains into the information recommendation model to be trained to obtain the predicted recommendation degree of the sample information; based on the difference between the predicted recommendation degree and the actual recommendation degree, a loss value is calculated; when the loss value is greater than or equal to a preset threshold, the model parameters of the first processing model and the second processing model in the information recommendation model to be trained are adjusted according to the loss value to obtain an adjusted information recommendation model, and the adjusted information recommendation model is repeatedly trained until the loss value obtained according to the trained information recommendation model is less than the preset threshold.

[0159] In one embodiment, the determination module 540 is further used to filter out target information with a recommendation degree greater than a preset recommendation degree from the information to be recommended; and recommend the target information to the corresponding user.

[0160] Each module in the above information recommendation device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each module.

[0161] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as recommendation degree. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an information recommendation method is implemented.

[0162] Those skilled in the art will understand that Figure 6The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0163] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0164] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0165] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0166] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0167] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0168] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0169] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. An information recommendation method, characterized in that: The method comprises: Acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features; According to the feature domains corresponding to the information features, determining mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains; the information feature combination includes two information features; Inputting the mutual information value between the information feature and the feature domains corresponding to the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended; Determining a recommendation result of the information to be recommended according to the recommendation degree; The step of inputting the mutual information value between the feature domains corresponding to the information feature and the information feature combination of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended includes: Obtaining an information feature vector of the information feature through a first processing model in the pre-trained information recommendation model; Processing the N-order information feature combination according to the information feature vector to obtain a first output result; N is greater than or equal to 3; Processing the information feature, the information feature combination, the information feature vector and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains through the second processing model in the pre-trained information recommendation model to obtain a second output result; Performing a weighted summation on the first output result and the second output result to obtain a recommendation degree of the information to be recommended; The second processing model in the pre-trained information recommendation model processes the information feature, the information feature combination, the information feature vector and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains to obtain a second output result, including: Determine, by means of a second processing model in the pre-trained information recommendation model, a weight corresponding to the information feature, and determine, as a weight corresponding to the information feature combination, a product of an inner product of an information feature vector corresponding to an information feature in the information feature combination and a mutual information value between feature domains corresponding to the information feature combination; Obtaining a first product result between the information feature and the weight corresponding to the information feature, and a second product result between the information feature combination and the weight corresponding to the information feature combination; The first multiplication result and the second multiplication result are added to obtain the second output result.

2. The method according to claim 1, characterized in that: The determining, according to the feature domains corresponding to the information features, mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains comprises: Determining information feature combinations of different feature domains according to feature domains corresponding to the information features; Counting the individual occurrence frequency of each feature domain corresponding to the information feature combination of the different feature domains, and the common occurrence frequency of the corresponding feature domains; According to the individual occurrence frequency of each feature domain corresponding to the information feature combination of the different feature domains and the common occurrence frequency of the corresponding feature domains, the mutual information value between the feature domains corresponding to the information feature combination of the different feature domains is obtained.

3. The method according to claim 2, characterized in that The obtaining, according to the individual occurrence frequency of each feature domain corresponding to the information feature combination of the different feature domains and the common occurrence frequency of the corresponding feature domains, the mutual information value between the feature domains corresponding to the information feature combination of the different feature domains comprises: Determine the product of the individual occurrence frequencies of each feature domain corresponding to the information feature combination of the different feature domains; According to the ratio between the common occurrence frequency of the feature domains corresponding to the information feature combinations of the different feature domains and the product, query the corresponding relationship between the ratio and the feature domain association degree to obtain the association degree between the feature domains corresponding to the information feature combinations of the different feature domains; The correlation degree between the feature domains corresponding to the information feature combinations of the different feature domains is determined as the mutual information value between the feature domains corresponding to the information feature combinations of the different feature domains.

4. The method according to claim 1, characterized in that: The pre-trained information recommendation model is trained in the following way: Acquire training data; the training data includes sample information features of sample information, mutual information values ​​between feature domains corresponding to combinations of sample information features of different feature domains, and actual recommendation degrees of the sample information; According to the training data, jointly training the first processing model and the second processing model in the information recommendation model to be trained until a training end condition is reached; The information recommendation model that meets the training end condition is determined as the pre-trained information recommendation model.

5. The method according to claim 4, characterized in that The step of jointly training the first processing model and the second processing model in the information recommendation model to be trained according to the training data until a training end condition is reached includes: Inputting the mutual information value between the feature domains corresponding to the sample information feature and the combination of the sample information features of the different feature domains into the information recommendation model to be trained to obtain the predicted recommendation degree of the sample information; Calculating a loss value according to a difference between the predicted recommendation degree and the actual recommendation degree; When the loss value is greater than or equal to a preset threshold, the model parameters of the first processing model and the second processing model in the information recommendation model to be trained are adjusted according to the loss value to obtain an adjusted information recommendation model, and the adjusted information recommendation model is repeatedly trained until the loss value obtained according to the trained information recommendation model is less than the preset threshold.

6. The method according to any one of claims 1 to 5, characterized in that: Determining the recommendation result of the information to be recommended according to the recommendation degree includes: Filtering out target information whose recommendation degree is greater than a preset recommendation degree from the information to be recommended; The target information is recommended to the corresponding user.

7. An information recommendation device, characterized in that: The device comprises: An acquisition module, used to acquire information features of the information to be recommended, perform domain classification on the information features, and obtain feature domains corresponding to the information features; A statistical module, for determining, according to the feature domain corresponding to the information feature, mutual information values ​​between feature domains corresponding to information feature combinations of different feature domains; the information feature combination includes two information features; A prediction module, used for inputting the mutual information value between the feature domains corresponding to the information feature and the combination of information features of different feature domains into a pre-trained information recommendation model to obtain the recommendation degree of the information to be recommended; A determination module, used for determining a recommendation result of the information to be recommended according to the recommendation degree; Wherein, the prediction module is further used to obtain the information feature vector of the information feature through the first processing model in the pre-trained information recommendation model; process the N-order information feature combination according to the information feature vector to obtain a first output result; N is greater than or equal to 3; process the information feature, the information feature combination, the information feature vector and the mutual information value between the feature domains corresponding to the information feature combination of different feature domains through the second processing model in the pre-trained information recommendation model to obtain a second output result; perform weighted summation on the first output result and the second output result to obtain the recommendation degree of the information to be recommended; The prediction module is also used to determine the weight corresponding to the information feature through the second processing model in the pre-trained information recommendation model, and determine the product of the inner product of the information feature vector corresponding to the information feature in the information feature combination and the mutual information value between the feature domains corresponding to the information feature combination as the weight corresponding to the information feature combination; obtain a first product result between the information feature and the weight corresponding to the information feature, and a second product result between the information feature combination and the weight corresponding to the information feature combination; add the first product result and the second product result to obtain the second output result.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

Images