Method, device, storage medium and electronic device for providing information
By extracting and associating weighted features and combining with multi-sub network structure, the accuracy of personalized recommendation technology is improved, and the problem of low user preference prediction accuracy in the prior art is solved, achieving more efficient information provision.
Patent Information
- Application Number
- CN202011290242.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-11-17
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2040-11-17
AI Technical Summary
In the existing personalized recommendation technology, the machine learning model based on user behavior has a low accuracy rate when predicting user preference information, making it difficult to effectively provide information in line with user preferences.
By extracting the features in the target information and behavioral data, using the pre-trained information provision model to determine the information provision method, combining the behavioral characteristics after association weighting, outputting the target recommendation degree, and providing information to the user based on the recommendation degree, adopting a multi-sub network structure and association weighting mechanisms improve recommendation accuracy.
The accuracy of user preference prediction is improved, and the user's historical behavior is given more closely related behaviors through the association weighting mechanism, which enhances the ability to predict target information.
Smart Images

Figure CN112417275B_ABST
Abstract
Description
Technical Field
[0001] The present specification relates to the field of computer technology, and in particular to a method for providing information, a device storage medium, and an electronic device. Background Art
[0002] With the popularization of the Internet, more and more information is produced and released. The explosive growth of information makes it increasingly difficult for users to obtain information that meets their preferences. Personalized recommendation technology has become an indispensable basic technology in the Internet field.
[0003] In existing personalized recommendation technologies, there are some machine learning models that predict user preference information. Preference predictions are often made based on user profiles obtained according to user behavior. However, the accuracy of preference predictions obtained using this method is low. Summary of the invention
[0004] This specification provides an information providing method and device to partially solve the above-mentioned problems existing in the prior art.
[0005] This manual adopts the following technical solutions:
[0006] This specification provides an information providing method, including:
[0007] Determine each target information to be provided, and each behavior data of the user;
[0008] Extracting information features from each target information; extracting behavior features from each behavior data; and inputting each information feature and each behavior feature into a pre-trained information provision model, wherein the information provision model is used to determine the information provision method of each target information;
[0009] For each information feature input, the association weight between the information feature and each behavior feature is determined through the information provision model, and the corresponding behavior features are weighted by the association weight to determine each behavior feature after the association weight;
[0010] Determining the target recommendation degree of the target information output by the information provision model according to each behavior feature after association weighting;
[0011] According to the target recommendation degree of each target information, an information providing method of each target information is determined, and at least part of the target information is provided to the user in the information providing method.
[0012] Optionally, the information feature extracted from each target information is used as the first information feature;
[0013] Before determining the target recommendation degree of the target information output by the information providing model, the method further includes:
[0014] Determine relevant information corresponding to the target information;
[0015] Extracting information features from the relevant information corresponding to the target information as second information features;
[0016] Input each information feature and each behavior feature into the pre-trained information provision model, including:
[0017] Each first information feature, each second information feature, and each behavior feature are input into a pre-trained information provision model.
[0018] Optionally, determining the association weights between the information feature and each behavior feature through the information provision model specifically includes:
[0019] For the input first information feature, determine the first association weight between the first information feature and each behavior feature through the information provision model, and associate and weight the corresponding behavior features with the first association weight to determine each behavior feature after association and weighting;
[0020] For the input second information feature, the second association weight between the second information feature and each behavior feature is determined through the information provision model, and the corresponding behavior features are associated and weighted with the second association weight to determine each behavior feature after association weighting.
[0021] Optionally, the information provision model includes at least two sub-networks;
[0022] Inputting each first information feature, each second information feature, and each behavior feature into a pre-trained information provision model specifically includes:
[0023] Input each first information feature and each first behavior feature into the first sub-network; input each second information feature and each second behavior feature into the second sub-network; wherein the first behavior feature is a behavior feature extracted from the behavior data with the target information as the object, and the second behavior feature is a behavior feature extracted from the behavior data with the related information as the object.
[0024] Optionally, determining the association weights between the information feature and each behavior feature through the information provision model, and performing association weighting on the corresponding behavior features with the association weights to determine each behavior feature after association weighting specifically includes:
[0025] For the input first information feature, determine the first association weight between the first information feature and each first behavior feature through the information provision model, and associate and weight the corresponding first behavior features with the first association weight to determine each first behavior feature after association and weighting;
[0026] For the input second information feature, the second association weight between the second information feature and each second behavior feature is determined through the information provision model, and the corresponding second behavior features are associated and weighted with the second association weight to determine each second behavior feature after association weighting.
[0027] Optionally, before determining the target recommendation degree of the target information output by the information providing model, the method further includes:
[0028] Determine the first comprehensive feature of the target information according to each second behavior feature after association weighting;
[0029] Determining a second comprehensive feature of the target information according to each of the first behavior features and the first comprehensive feature of the target information;
[0030] A target recommendation degree of the target information is determined according to the second comprehensive feature of the target information.
[0031] Optionally, providing at least part of the target information to the user in the information providing manner specifically includes:
[0032] At least part of the target information and at least part of the related information corresponding to each provided target information are provided to the user in the information providing manner.
[0033] Optionally, the target information specifically includes: information of the merchant;
[0034] The relevant information corresponding to the target information specifically includes: information about commodities provided by merchants.
[0035] Optionally, the information providing model is pre-trained, specifically including:
[0036] Determine an information provision model to be trained and training samples for training the information provision model, wherein the training samples include: each target information and each user's historical behavior data;
[0037] Extracting information features from each target information; extracting behavior features from each behavior data, and inputting them into the information providing model to be trained;
[0038] For each information feature, determine the association weight of the information feature to each behavior feature, and perform association weighting on the corresponding behavior features according to the association weights to determine each behavior feature after association weighting, and determine the target recommendation degree of the target information corresponding to the information feature according to each behavior feature after association weighting;
[0039] Determining labeling information of each target information according to the behavior data;
[0040] The loss of the information providing model is determined according to the target recommendation degree and the labeling information of each target information, and the information providing model is trained with the goal of minimizing the loss.
[0041] This specification provides an information providing device, including:
[0042] The acquisition module is used to determine each target information to be provided and each behavior data of the user;
[0043] A feature input module is used to extract information features from each target information; extract behavior features from each behavior data; and input each information feature and each behavior feature into a pre-trained information provision model, wherein the information provision model is used to determine the information provision method of each target information;
[0044] A feature weighting module, for determining, for each information feature input, an association weight between the information feature and each behavior feature through the information provision model, and performing association weighting on the corresponding behavior feature with the association weight to determine each behavior feature after association weighting;
[0045] A target recommendation module, used to determine the target recommendation degree of the target information output by the information provision model according to each behavior feature after association weighting;
[0046] The information providing module is used to determine the information providing mode of each target information according to the target recommendation degree of each target information, and provide at least part of the target information to the user in the information providing mode.
[0047] This specification provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the above-mentioned information providing method is implemented.
[0048] This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned information providing method when executing the program.
[0049] At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:
[0050] In the information providing method provided in this specification, the association weight between any information feature and any behavior feature is determined according to the degree of association between the user's behavior and the target information to be provided, and the behavior feature is associated and weighted with the association weight, so that different user historical behaviors are assigned different association weights according to the correlation between the behavior and the target information, thereby giving user behaviors that are more closely related to the target information a greater voice in predicting the target information. Compared with the case of non-association weights, the accuracy of predicting users' preferences for target information based on each user's behavior is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] The drawings described herein are used to provide a further understanding of this specification and constitute a part of this specification. The illustrative embodiments and descriptions of this specification are used to explain this specification and do not constitute an improper limitation on this specification. In the drawings:
[0052] Figure 1 A flowchart of a model training method in this specification;
[0053] Figure 2 A flowchart of another model training method in this specification;
[0054] Figure 3 A schematic diagram of the model structure is provided for one type of information in this manual;
[0055] Figure 4 A flowchart of an information providing method in this manual;
[0056] Figures 5A to 5C A schematic diagram of a method for providing three types of relevant information in this manual;
[0057] Figure 6 A schematic diagram of an information providing device provided in this specification;
[0058] Figure 7 The corresponding Figure 4 Schematic diagram of electronic equipment. DETAILED DESCRIPTION
[0059] In order to make the purpose, technical solutions and advantages of this specification more clear, the technical solutions of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.
[0060] The information provision method described in this specification is intended to determine an information provision plan and provide it to users based on the collected behavioral data that can characterize the user's information preferences. In other words, it is to make a selection from a large amount of information and provide the selected information to the user in a specific manner, that is, personalized information recommendation.
[0061] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.
[0062] It should be noted that since each embodiment of the information provision recorded in this specification is based on the machine learning model used, that is, an improvement of the information provision model in this specification, in this context, this specification also provides an embodiment of training the information provision model.
[0063] Embodiment 1:
[0064] Figure 1 The model training process diagram provided for the embodiments of this specification includes:
[0065] S100: Determine an information provision model to be trained and training samples for training the information provision model, wherein the training samples include: each target information and each user's historical behavior data.
[0066] The information providing model to be trained (hereinafter referred to as the model to be trained) in this specification has the following goal after training, that is, the expected function to be achieved: from each target information input, at least part of the target information is determined as the information to be provided, and provided to the user in the determined provision method. For example, it can be the screening of each target information and the sorting of the screening results. Of course, this specification does not limit the way in which the screened target information is provided, and the target information can also be provided in an unordered manner.
[0067] The training samples used to train the model to be trained in this specification include: each target information and each user's historical behavior data. From the above description, it can be seen that the target information is used as the input of the model to be trained, and the model to be trained is calculated and screened in a preset manner according to the historical behavior data of each user, and the screened target information is also used as the output of the model to be trained.
[0068] Depending on the usage scenario, various types of information can be used as target information. Generally speaking, in the field of e-commerce, the information of merchants or products can be used as target information; while for content platforms, the information of creators or their works can be used as target information.
[0069] It should be noted that since different users often do not have the same preferences, although the training samples include historical behavior data of several users, for a single training (i.e., one input and output of the model to be trained), only the historical behavior data of a single user is selected as the sample, and the target information in the sample is labeled according to the behavior data of the single user, and the preference prediction result for the user is output.
[0070] Before training, the information provision model to be trained can be an initialized model, in which the parameters of the model are pre-set or randomly generated; or it can be a model that has been trained in the past but failed to meet the use requirements. In this case, the model can be trained again. In other words, any information provision model that needs to be trained to improve the model information provision performance can be the information provision model to be trained in this specification.
[0071] S102: extracting information features from each target information; extracting behavior features from each behavior data, and inputting them into the information providing model to be trained.
[0072] This specification does not limit the method of extracting information features from each target information and extracting behavior features from each behavior data, so any existing feature extraction method can be selected, and the extracted information features and behavior features can be input into the information provision model to be trained.
[0073] It is worth noting that this description does not limit the correspondence between each target information / behavior data before extraction and an information feature / behavior feature after extraction. It is also possible to extract each behavior data into a behavior feature, etc. For the convenience of description, this description takes the example that each target information / behavior data before and after feature extraction corresponds to an information feature / behavior feature.
[0074] S104: For each information feature, determine the association weight of the information feature with each behavior feature.
[0075] Generally speaking, each user can have multiple behaviors, which is reflected in the embodiments of this specification, that is, a large amount of behavior data can be obtained for each user, but not every behavior data will affect the user's preference for target information. For example, when the target information and the objects of user behavior are both commodities, for a user, the behavior data obtained include behavior data for commodities such as potato chips, earrings, and cat food. When the target commodity is a cat bed, generally, the user's preference for cat beds is less correlated with his behavior data for potato chips and earrings, but is closely correlated with his behavior data for cat food.
[0076] Based on this, for each information feature, according to the degree of association between the information feature and each behavior data, an association weight for the information feature is determined for each behavior data to characterize the degree of association between the behavior data and the information feature, wherein the larger the association weight, the closer the association between the behavior data and the information feature, and conversely, the smaller the association weight, the more distant the association between the behavior data and the information feature. In one embodiment of the present specification, an attention mechanism can be used to determine the association weight between the information feature and the behavior feature by allocating attention weights.
[0077] It can be understood that not only for one information feature, the degree of association between different behavior features and the information feature is different, and for different information features, the degree of association between the same behavior feature and different information features is also different. In other words, the degree of association between any information feature and any behavior feature is different, and the association weights between them are also different.
[0078] S106: performing association weighting on the corresponding behavior features according to the association weights to determine each association weighted behavior feature, and determining a target recommendation degree of the target information corresponding to the information feature according to each association weighted behavior feature.
[0079] It is worth noting that the target recommendation degree of the target information output by the model to be trained can express the user's preference for the target information as believed by the model to be trained. Generally speaking, in the Internet field, the prediction of click-through rate is often used to represent the prediction of user preference. In this situation, the predicted click-through rate of the target information can be determined based on the target recommendation degree of the target information. In other words, the target recommendation degree of the target information is the predicted click-through rate of the target information.
[0080] After determining the target recommendation degree of each target information, the loss of the model to be trained can be directly determined based on the labeling information and the target recommendation degree of each target information; or the target information to be provided can be determined based on the target recommendation degree of each target information, and the loss of the model to be trained can be determined based on the labeling information and the target recommendation degree of each target information to be provided.
[0081] When each behavior data is extracted as a multidimensional behavior feature, it can be understood that the association weighting for the behavior feature can also be a multidimensional vector (or matrix) composed of multiple association weights. Those skilled in the art can understand that when the behavior feature is association weighted by the multidimensional vector composed of the association weights, it is not that the behavior feature as a whole is association weighted with the same fixed association weight, but that each association weight in the multidimensional vector composed of the association weights is weighted for the corresponding dimension of the behavior feature, such as the dot product between two matrices.
[0082] S108: Determine labeling information for each target information according to the behavior data.
[0083] This specification only exemplifies one marking method, and any existing marking method can be used to mark the target information, and other marking methods will not be described in detail.
[0084] The user's historical behavior data used in this training includes the user's historical preference for each target information (such as click-through rate), and based on the target recommendation degree of each target information, the target recommendation degree reflects the user's preference for each target information as believed by the model to be trained (such as click-through rate prediction results). In this scenario, the above behavior data can be used as a label for the target information corresponding to the behavior data.
[0085] S110: Determine the loss of the information providing model according to the target recommendation degree and the labeling information of each target information, and train the information providing model with the goal of minimizing the loss.
[0086] According to the prediction results output by the model to be trained and the labels of the corresponding samples, the loss of the model to be trained is determined. The smaller the loss, the closer the prediction result output by the model to be trained is to the label, and vice versa, the greater the difference between the prediction result and the label. The gradient can be determined based on the loss, and the gradient descent algorithm can be used to adjust the model parameters in the model to be trained. When the number of model parameter adjustments reaches a preset number threshold, and / or the loss is less than a preset loss threshold, the model to be trained is considered to be a trained information providing model.
[0087] Continuing with the above example, for each target information, the actual click rate of the user for the target information can be used as the label of the target information. According to the label information and the click rate prediction result of the model to be trained for the target information (i.e., the target recommendation degree), the loss of the model to be trained is determined, and the gradient descent method is used to adjust each model parameter in the model to be trained until the number of adjustments of the model parameters reaches a preset number threshold, and / or the loss is less than a preset loss threshold.
[0088] The above describes the process of single training of the model to be trained, where single training means that the model to be trained is input once and output once, and the loss is determined once based on the labeled information. But generally speaking, a model that can achieve the required prediction accuracy often requires multiple trainings. For the purpose of enriching the training samples, the behavior data of the same user is not selected as the training sample for each training of the model to be trained in this specification.
[0089] Furthermore, the user's preference for the target information often depends not only on the target information itself, but also on other information related to the target information, which may also contain the user's preference for the target information.
[0090] Continuing with the above example, when the target information is a merchant, the user's preference for the products sold by the merchant is generally similar to the user's preference for the merchant; and when the target information is a creator, such as a director, whether the user prefers the director depends largely on the film and television works directed by the director. Therefore, in this manual, relevant information related to the target information can be used to train the model to be trained.
[0091] It can be seen that there is a corresponding relationship between the relevant information and the target information, that is, the relevant information can only exist as relevant information of the target information, and can only affect the target information with which the corresponding relationship exists. However, how many relevant information corresponds to one target information, and whether one relevant information can correspond to several target information, are not limited in the embodiments of this specification. For the convenience of explanation, the following only takes the example that one target information can correspond to several relevant information, but one relevant information can only correspond to one target information.
[0092] In this situation, in order to make a distinction, the information feature extracted from the target information is used as the first information feature; the information feature extracted from the relevant information is used as the second information feature, and both the first information feature and the second information feature are input into the model to be trained.
[0093] For each information feature (including the first information feature and the second information feature), the association weight between the information feature and each behavior feature is determined, and each behavior feature is associated and weighted based on the determined association weight. Specifically, the association weight determined between the first information feature and the behavior feature can be used as the first association weight; the association weight determined between the second information feature and the behavior feature can be used as the second association weight.
[0094] Behavioral features can be associated and weighted with each information feature without distinction. However, when the behavioral object of the behavioral data can be determined, each behavioral feature extracted from the behavioral data can be determined as the first behavioral feature or the second behavioral feature according to its corresponding behavioral data based on the behavioral object of each behavioral data. Specifically, the first behavioral feature can be the behavioral feature extracted from the behavioral data with the target information as the object, and the second behavioral feature can be the behavioral feature extracted from the behavioral data with the related information as the object.
[0095] In this context, Figure 1The process of determining the association weights between information features and behavior features in steps S104 to S106, and determining the target recommendation degree of the target information according to the associated weighted behavior features, can be as follows: Figure 2 shown.
[0096] Figure 2 The embodiment of this specification provides a process for determining the association weight between information features and behavior features, and determining the target recommendation degree of target information according to each behavior feature after association weighting, which specifically includes the following steps:
[0097] S200: Determine each first behavior feature after association weighting.
[0098] Determine the first association weight between the first information feature (the information feature extracted from the target information) and each first behavior feature (that is, the corresponding behavior data is the behavior feature with the target information as the object), and associate and weight the corresponding first behavior features with the first association weight to determine each first behavior feature after association weighting.
[0099] S202: Determine each second behavior feature after association weighting.
[0100] Determine the second association weight between each second information feature (information feature extracted from the target information) and each second behavior feature (i.e., the corresponding behavior data is a behavior feature with the relevant information as the object), and associate and weight the corresponding second behavior features with the second association weight to determine each second behavior feature after association weighting.
[0101] S204: Determine a target recommendation degree of the target information according to each first behavior feature after association weighting and each second behavior feature after association weighting.
[0102] Determine the first comprehensive feature of the target information based on the associated weighted second behavior features; determine the second comprehensive feature of the target information based on the first behavior features and the first comprehensive feature of the target information; determine the target recommendation degree of the target information based on the second comprehensive feature of the target information.
[0103] Among them, in step S204, each second behavioral feature after association weighting can use any existing algorithm to determine the first comprehensive feature of the target information, such as any pooling method; each first behavioral feature and the first comprehensive feature of the target information can use any existing algorithm to determine the second comprehensive feature of the objective function, such as a Concat function, etc. The embodiment of this specification will not elaborate on how to determine the first comprehensive feature of the target information and the second comprehensive feature of the objective function.
[0104] It is worth noting that in steps S200 to S202, as to how to use the model to be trained to determine the first behavioral features after association weighting and the second behavioral features after association weighting, it can be that for each information feature in turn, the association weight between the information feature and each behavioral feature is determined and the association weighting is performed, or for each information feature, the association weight between the information feature and each behavioral feature is calculated and associated weighting is performed in parallel in multiple threads of the computer. In addition, in one embodiment of the present specification, a method based on the modification of the model structure is provided, so that the first information feature and the second information feature are respectively calculated in different sub-networks. The association weights between the corresponding behavioral features are calculated and associated weighting is performed.
[0105] For example, the model to be trained includes at least two sub-networks, which can be distinguished as a first sub-network and a second sub-network, respectively. The first information feature and each first behavior feature are input into the first sub-network; each second information feature and each second behavior feature are input into the second sub-network.
[0106] In the first subnetwork, the first association weight between the first information feature and each first behavior feature is determined, and the corresponding first behavior features are associated and weighted with the first association weight to determine each first behavior feature after the association weighting; in the second subnetwork, for each second information feature, the second association weight between the second information feature and each second behavior feature is determined, and the corresponding second behavior features are associated and weighted with the second association weight to determine each second behavior feature after the association weighting.
[0107] That is to say, steps S200 to S202 are completed in such a manner that the two sub-networks respectively calculate the features corresponding to the target information and the related information.
[0108] In this scenario, step S204 may be to determine the first comprehensive feature of the target information based on the associated weighted second behavior features corresponding to the second information features output by the second subnetwork; determine the second comprehensive feature of the target information based on the associated weighted first behavior features output by the first subnetwork and the first comprehensive feature of the target information; and determine the target recommendation degree of the target information based on the second comprehensive feature of the target information.
[0109] The structure of the model to be trained can be as follows Figure 3As shown, in subnetwork 1, the association weight 1A between information feature 1 and behavior feature A is determined, and the behavior feature A is association-weighted with the association weight 1A to obtain the association-weighted behavior feature 1A; in subnetwork 2, the association weights 2B and 3B between information features 2 and 3 and behavior feature B are respectively determined, and the behavior feature B is association-weighted according to the association weights 2B and 3B to obtain the association-weighted behavior features 2B and 3B, and the first comprehensive feature of the target information is determined based on the association-weighted behavior features 2B and 3B, and the second comprehensive feature of the target information is determined based on the first comprehensive feature and the association-weighted behavior feature 1A.
[0110] Generally speaking, in addition to the target, related information, and behavioral data, there is other information that can determine the user's preference for the target information, such as the user's profile information, including gender, age, etc., which can also be extracted as features and used as Figure 3 The information feature 4 or the behavior feature C shown is involved in predicting the user's preference for the target information.
[0111] Furthermore, the target function of the model to be trained may also include, after determining the method for providing the target information, providing at least part of the relevant information corresponding to each target information provided.
[0112] This specification does not limit how to train the model to be trained to provide relevant information corresponding to each target information, but provides three methods as examples:
[0113] The first method is to determine the relevant information corresponding to each target information provided, and determine the prediction results of the relevant information according to the second behavior features after the association weighting determined in the second step for the relevant information.
[0114] The second type: for each target information determined to be provided and each related information corresponding to each target information, the identities of the target information and the related information are exchanged, that is, for each related information, the related information is used as the target information, and the target information corresponding to the related information is used as the related information, and based on the exchanged target information and related information, each behavioral feature after associated weighting is determined, and based on each behavioral feature after each associated weighting, the target recommendation degree of each exchanged target information (that is, the related information before the exchange) is determined.
[0115] For each target information to be provided before the exchange, the prediction results of the relevant information are determined according to the target recommendation degrees of the relevant information corresponding to the target information and the target recommendation degrees of the relevant information.
[0116] The third method is to input the second comprehensive feature of the target information determined by the above method into another fully connected layer that is different from the fully connected layer of the model to be trained, so as to obtain a new model to be trained, and the target function of the new model to be trained is to determine the prediction results of each relevant information.
[0117] Based on the prediction results of the relevant information determined by any of the above methods and the labeling information of the relevant information determined by the behavioral data, the loss of the model to be trained is determined, and with the goal of minimizing the loss, the model parameters in the model to be trained are adjusted.
[0118] In addition, this specification also provides a training method, which determines the loss of the model to be trained based on the prediction results of the model to be adjusted for the target information and related information, as well as the labeling information of the target information and behavior information determined according to the behavior data, and adjusts each model parameter in the model to be trained with the goal of minimizing the loss, thereby simultaneously improving the prediction accuracy of the model to be trained for the target information and related information.
[0119] Based on the information providing model trained in the above-mentioned embodiment 1, this specification provides a corresponding information providing method, such as embodiment 2.
[0120] Embodiment 2:
[0121] Figure 4 The information provided in the embodiments of this specification provides a flow chart, including:
[0122] S400: Determine various target information to be provided and various behavior data of the user.
[0123] Determine the usage scenario of the information provision model, and use various information that may be provided to the user as target information. Determine the user who receives the information provided by the information provision model, and determine various behavior data of the user.
[0124] In the embodiments of this specification, when the information to be provided and the method of providing are determined each time, the information provision is deemed to be completed once. For the convenience of description, only the embodiments in the process of information provision are described below. Generally speaking, information is only provided to one user each time. Therefore, each information provision is only provided based on the behavioral data of the user who wants to receive the information.
[0125] S402: extracting information features from each target information; extracting behavior features from each behavior data; and inputting each information feature and each behavior feature into a pre-trained information provision model.
[0126] The information provision model is based on Figure 1The model trained by the model training method shown is used to determine the information provision method of each target information.
[0127] Adopt and Figure 1 The same method is used to extract information features from each target information; and behavior features are extracted from each behavior data. It should be noted that this specification does not limit the one-to-one correspondence between target information and information features, nor does it limit the one-to-one correspondence between behavior data and behavior features. Those skilled in the art can understand that grouping each behavior data into a behavior sequence and extracting corresponding behavior features from the behavior sequence is a commonly used feature extraction method. However, for ease of explanation, this specification will take the one-to-one correspondence between target information and information features, and the one-to-one correspondence between behavior data and behavior features as an example for explanation.
[0128] S404: For each input information feature, determine the association weight between the information feature and each behavior feature through the information provision model.
[0129] pass Figure 1 The information provision model trained by the method shown determines the association weights between each information feature and each behavior feature.
[0130] S406: performing association weighting on the corresponding behavior features using the association weights to determine each behavior feature after association weighting;
[0131] According to each determined association weight, the corresponding behavior feature is association-weighted. It should be noted that for each behavior feature, if the behavior feature and several information features have determined several association weights, the behavior feature needs to be association-weighted several times.
[0132] However, in this situation, the information feature only includes the information feature of the target information, and it can be understood that the behavior feature is only once associated with the target information using the associated weight determined by the information feature.
[0133] S408: Determine the target recommendation degree of the target information output by the information provision model according to each behavior feature after association weighting;
[0134] This specification does not limit how to determine the target recommendation degree based on the associated weighted behavioral features. It can be determined directly based on the associated weighted behavioral features with the largest associated weight of each information feature, or it can be determined by summing up the associated weighted behavioral features of each information feature. This specification will not elaborate on how to determine the target recommendation degree based on the associated weighted behavioral features.
[0135] S410: Determine an information providing method for each target information according to the target recommendation degree of each target information, and provide at least part of the target information to the user in the information providing method.
[0136] In the embodiments of this specification, there is no limitation on how to determine the information provision method of each target information based on the target recommendation degree of each target information. As an example, the ranking of each target information can be determined based on the target recommendation degree, and the top N target information are selected and provided to the user in order of the ranking of each target recommendation degree, wherein N is a preset ranking threshold.
[0137] Generally speaking, since the prediction of click-through rate is often used to represent the prediction of user preference in the Internet field, in this situation, the predicted click-through rate of the target information can be determined based on the recommendation degree of the target information. In other words, the recommendation degree of the target information is the predicted click-through rate of the target information. Therefore, it can be considered that the higher the target recommendation degree of the target information, the more confident it is that the user who receives the target information will click on the target information, thereby achieving personalized information provision.
[0138] Above Figure 4 It can be seen from the described method that this specification proposes an information providing method, which determines the association weight between any information feature and any behavior feature according to the degree of association between the user's behavior and the target information to be provided, and associates and weights the behavior features with the association weights, so as to achieve different user historical behaviors. According to the correlation between the behavior and the target information, different association weights are assigned, thereby giving user behaviors that are more closely related to the target information a greater voice in predicting the target information. Compared with the case of non-association weights, the accuracy of predicting users' preferences for target information based on each user's behavior is improved.
[0139] This manual Figure 4 The method of determining the target recommendation degree of a target information by using the information provision model is exemplified in the following. However, since there are several target information to be provided, it is obviously necessary to calculate the target recommendation degree of each target information. However, since the same information provision model can only calculate the target recommendation degree of one target information at a time, it can be used in sequence. Figure 4 The method shown calculates the target recommendation degree of each target information, and the information provision model may also be run in multiple threads at the same time, that is, the target recommendation degrees of multiple target information are calculated in multiple threads at the same time and in parallel.
[0140] Since the information providing model used in the information providing method provided in this specification has a model structure as follows Figure 3 As shown, therefore, based on this model, this specification also provides the following information providing method.
[0141] In addition to determining the target information to be provided based on the target information itself, it is also possible to determine the target information based on the relevant information related to the target information, using information features extracted from each target information as first information features; determining the relevant information corresponding to the target information; extracting information features from the relevant information corresponding to the target information as second information features; and inputting each first information feature, each second information feature, and each behavioral feature into a pre-trained information provision model.
[0142] For the purpose of distinction, the association weight determined by the first information feature and the behavior feature can be used as the first association weight; the association weight determined by the second information feature and the behavior feature can be used as the second association weight. For the input first information feature, the first association weight between the first information feature and each behavior feature is determined by the information provision model, and the corresponding behavior features are weighted by the first association weight to determine each behavior feature after the association weighting; for the input second information feature, the second association weight between the second information feature and each behavior feature is determined by the information provision model, and the corresponding behavior features are weighted by the second association weight to determine each behavior feature after the association weighting.
[0143] The behavioral features can also be determined as the first behavioral model and the second behavioral model based on the behavioral data from which they are derived. The first behavioral feature is a behavioral feature extracted from the behavioral data with target information as the object, and the second behavioral feature is a behavioral feature extracted from the behavioral data with related information as the object.
[0144] based on Figure 3 The information provision model of the structure shown can use at least two subnetworks to respectively calculate the association weight between the first information feature and the first behavior feature, and the association weight between the second information feature and the second behavior feature, and input each first information feature and each first behavior feature into the first subnetwork; and input each second information feature and each second behavior feature into the second subnetwork. For the input first information feature, the first association weight between the first information feature and each first behavior feature is determined by the information provision model, and the corresponding first behavior feature is weighted by the first association weight to determine each first behavior feature after the association weighting; for the input second information feature, the second association weight between the second information feature and each second behavior feature is determined by the information provision model, and the corresponding second behavior feature is weighted by the second association weight to determine each second behavior feature after the association weighting.
[0145] After calculating the first behavior feature after association weighting and each second behavior feature after association weighting, the target recommendation degree of the target information can be calculated based on the first behavior feature after association weighting and each second behavior feature after association weighting, and the first comprehensive feature of the target information can be determined; the second comprehensive feature of the target information can be determined based on the first behavior features and the first comprehensive feature of the target information; and the target recommendation degree of the target information can be determined based on the second comprehensive feature of the target information.
[0146] Furthermore, after determining the method for providing the target information, for each target information provided, at least part of the relevant information corresponding to the target information is also provided.
[0147] The following takes the example where the target information is the merchant's information and the related information is the merchant's dish information to explain how the related information is provided.
[0148] Figures 5A to 5C Schematic diagrams of three information providing methods of this specification, wherein 5A is a schematic diagram of providing only relevant information (merchants), and merchants A to C are merchants selected by the method of the embodiment of this specification, and are provided to users in a manner of ranking according to the degree of recommendation; Figure 5B In addition to providing merchant information, a method of providing dish information is provided. Merchants A and B are two merchants selected using the method of the embodiment of this specification. Dishes 1 to 6 are dishes belonging to merchants A and B selected using the embodiment of this specification, wherein dishes 1 to 3 belong to merchant A, and dishes 4 to 6 belong to merchant B. Figure 5C Another way of providing information is to provide dish information on the basis of providing merchant information, wherein merchants A and B are two merchants selected using the method of the embodiment of this specification, dishes 1 to 2 are the dishes belonging to merchants A and B selected using the embodiment of this specification, and dishes 3 to 4 belong to merchant A.
[0149] Figure 5B and Figure 5C The difference is that Figure 5B It does not directly display the correspondence between merchants and dishes, but only displays the dishes; Figure 5C That is to say, the information provision method determined in this specification includes both the information to be provided and the provision method of each information: different information to be provided obviously means different information provision methods, and when the information to be provided is the same, different information provision methods are also provided in different ways.
[0150] So far, this specification provides an information providing method, which is based on the transformation of an information providing model. The information providing model can predict user preferences and provide target information and related information related to the target information based on the predicted preferences, and achieve a high accuracy rate.
[0151] The above is the information providing method provided in the embodiments of this specification. Based on the same idea, this specification also provides corresponding devices, storage media and electronic devices.
[0152] Figure 6 A schematic diagram of the structure of an information providing device provided in an embodiment of this specification, the device comprising:
[0153] The acquisition module 600 is used to determine various target information to be provided and various behavior data of the user;
[0154] The feature input module 602 is used to extract information features from each target information; extract behavior features from each behavior data; and input each information feature and each behavior feature into a pre-trained information provision model, wherein the information provision model is used to determine the information provision method of each target information;
[0155] The feature weighting module 604 is used to determine the association weight between each information feature input and each behavior feature through the information provision model, and to perform association weighting on the corresponding behavior feature with the association weight to determine each behavior feature after association weighting;
[0156] A target recommendation module 606 is used to determine the target recommendation degree of the target information output by the information provision model according to each behavior feature after association weighting;
[0157] The information providing module 608 is used to determine the information providing method of each target information according to the target recommendation degree of each target information, and provide at least part of the target information to the user in the information providing method.
[0158] The device also includes: a model training module 610, which is used to determine the information provision model to be trained, and training samples for training the information provision model, wherein the training samples include: each target information and each user's historical behavior data; extracting information features from each target information; extracting behavior features from each behavior data, and inputting them into the information provision model to be trained; for each information feature, determining the association weight of the information feature to each behavior feature, and according to the association weight, performing association weighting on the corresponding behavior features, determining each behavior feature after association weighting, and determining the target recommendation degree of the target information corresponding to the information feature according to each behavior feature after association weighting; determining the labeling information of each target information according to the behavior data; determining the loss of the information provision model according to the target recommendation degree and labeling information of each target information, and training the information provision model with the goal of minimizing the loss.
[0159] Optionally, the feature input module 602 is specifically used to use the information feature extracted from each target information as the first information feature; determine the relevant information corresponding to the target information; extract information features from the relevant information corresponding to the target information as the second information feature; and input each first information feature and each second information feature, as well as each behavioral feature into a pre-trained information provision model.
[0160] Optionally, the feature weighting module 604 is specifically used to, for a first information feature input, determine a first association weight between the first information feature and each behavior feature through the information provision model, and perform association weighting on the corresponding behavior features with the first association weight to determine each behavior feature after association weighting; for a second information feature input, determine a second association weight between the second information feature and each behavior feature through the information provision model, and perform association weighting on the corresponding behavior features with the second association weight to determine each behavior feature after association weighting.
[0161] Optionally, the feature weighting module 604 includes at least two sub-networks, specifically used to input each first information feature and each first behavior feature into a first sub-network; input each second information feature and each second behavior feature into a second sub-network; wherein the first behavior feature is a behavior feature extracted from the behavior data with the target information as the object, and the second behavior feature is a behavior feature extracted from the behavior data with the related information as the object;
[0162] The feature weighting module 604 is specifically used to, for the input first information feature, determine the first association weight between the first information feature and each first behavior feature through the information provision model, and perform association weighting on the corresponding first behavior features with the first association weight to determine each first behavior feature after association weighting; for the input second information feature, determine the second association weight between the second information feature and each second behavior feature through the information provision model, and perform association weighting on the corresponding second behavior features with the second association weight to determine each second behavior feature after association weighting.
[0163] Optionally, the target recommendation module 606 is specifically used to determine the first comprehensive feature of the target information based on the second behavioral features after association weighting; determine the second comprehensive feature of the target information based on the first behavioral features and the first comprehensive feature of the target information; and determine the target recommendation degree of the target information based on the second comprehensive feature of the target information.
[0164] Optionally, the information providing module 608 is specifically configured to provide the user with at least part of the target information and at least part of the related information corresponding to each provided target information in the information providing manner.
[0165] Optionally, the target information specifically includes: information about the merchant; and the relevant information corresponding to the target information specifically includes: information about goods provided by the merchant.
[0166] This specification also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 4 The information provided is provided by the method.
[0167] This manual also provides Figure 7 The schematic structure diagram of the electronic device shown in FIG. Figure 7 As mentioned above, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include other hardware required for the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 4 Of course, in addition to the software implementation, this specification does not exclude other implementations, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.
[0168] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.
[0169] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.
[0170] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
[0171] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.
[0172] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0173] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0174] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0175] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0176] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0177] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.
[0178] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
[0179] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.
[0180] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0181] This specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.
[0182] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0183] The above description is only an embodiment of the present specification and is not intended to limit the present specification. For those skilled in the art, the present specification may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification shall be included in the scope of the claims of the present specification.
Claims
1. A method for providing information, characterized in that: include: Determine each target information to be provided, and each behavior data of the user; Extract information features from each target information; Extracting behavioral features from each behavioral data; Each information feature and each behavior feature is input into a pre-trained information provision model, and the information provision model is used to determine the information provision method of each target information; the information provision model includes at least two sub-networks; each first information feature and each second information feature, as well as each behavior feature, are input into the pre-trained information provision model, specifically including: each first information feature and each first behavior feature are input into the first sub-network; each second information feature and each second behavior feature are input into the second sub-network; wherein the first behavior feature is a behavior feature extracted from the behavior data with the target information as the object, and the second behavior feature is a behavior feature extracted from the behavior data with the related information as the object; For each information feature input, the association weight between the information feature and each behavior feature is determined through the information provision model, and the corresponding behavior features are weighted by the association weight to determine each behavior feature after the association weight; Determining the target recommendation degree of the target information output by the information provision model according to each behavior feature after association weighting; According to the target recommendation degree of each target information, an information providing method of each target information is determined, and at least part of the target information is provided to the user in the information providing method.
2. The method according to claim 1, characterized in that The information feature extracted from each target information is used as the first information feature; Before determining the target recommendation degree of the target information output by the information providing model, the method further includes: Determine relevant information corresponding to the target information; Extracting information features from the relevant information corresponding to the target information as second information features; Input each information feature and each behavior feature into the pre-trained information provision model, including: Each first information feature, each second information feature, and each behavior feature are input into a pre-trained information provision model.
3. The method according to claim 2, characterized in that The association weights between the information feature and each behavior feature are determined by the information provision model, specifically including: For the input first information feature, determine the first association weight between the first information feature and each behavior feature through the information provision model, and associate and weight the corresponding behavior features with the first association weight to determine each behavior feature after association and weighting; For the input second information feature, the second association weight between the second information feature and each behavior feature is determined through the information provision model, and the corresponding behavior features are associated and weighted with the second association weight to determine each behavior feature after association weighting.
4. The method according to claim 1, characterized in that Determining the association weights between the information feature and each behavior feature through the information provision model, and weighting the corresponding behavior features with the association weights to determine each behavior feature after the association weights, specifically including: For the input first information feature, determine the first association weight between the first information feature and each first behavior feature through the information provision model, and associate and weight the corresponding first behavior features with the first association weight to determine each first behavior feature after association and weighting; For the input second information feature, the second association weight between the second information feature and each second behavior feature is determined through the information provision model, and the corresponding second behavior features are associated and weighted with the second association weight to determine each second behavior feature after association weighting.
5. The method according to claim 1, characterized in that Before determining the target recommendation degree of the target information output by the information providing model, the method further includes: Determine the first comprehensive feature of the target information according to each second behavior feature after association weighting; Determining a second comprehensive feature of the target information according to each of the first behavior features and the first comprehensive feature of the target information; A target recommendation degree of the target information is determined according to the second comprehensive feature of the target information.
6. The method according to claim 2, characterized in that Providing at least part of the target information to the user in the information providing manner specifically includes: At least part of the target information and at least part of the related information corresponding to each provided target information are provided to the user in the information providing manner.
7. The method according to claim 6, characterized in that The target information specifically includes: merchant information; The relevant information corresponding to the target information specifically includes: information about commodities provided by merchants.
8. The method according to any one of claims 1 to 7, characterized in that: Pre-training information providing model, including: Determine an information provision model to be trained and training samples for training the information provision model, wherein the training samples include: each target information and each user's historical behavior data; Extracting information features from each target information; extracting behavior features from each behavior data, and inputting them into the information providing model to be trained; For each information feature, determine the association weight of the information feature to each behavior feature, and perform association weighting on the corresponding behavior features according to the association weights to determine each behavior feature after association weighting, and determine the target recommendation degree of the target information corresponding to the information feature according to each behavior feature after association weighting; Determining labeling information of each target information according to the behavior data; The loss of the information providing model is determined according to the target recommendation degree and the labeling information of each target information, and the information providing model is trained with the goal of minimizing the loss.
9. An information providing device, characterized in that: include: The acquisition module is used to determine each target information to be provided and each behavior data of the user; A feature input module is used to extract information features from each target information; Extracting behavioral features from each behavioral data; Each information feature and each behavior feature is input into a pre-trained information provision model, and the information provision model is used to determine the information provision method of each target information; the information provision model includes at least two sub-networks; each first information feature and each second information feature, as well as each behavior feature, are input into the pre-trained information provision model, specifically including: each first information feature and each first behavior feature are input into the first sub-network; each second information feature and each second behavior feature are input into the second sub-network; wherein the first behavior feature is a behavior feature extracted from the behavior data with the target information as the object, and the second behavior feature is a behavior feature extracted from the behavior data with the related information as the object; A feature weighting module, for determining, for each information feature input, an association weight between the information feature and each behavior feature through the information provision model, and performing association weighting on the corresponding behavior feature with the association weight to determine each behavior feature after association weighting; A target recommendation module, used to determine the target recommendation degree of the target information output by the information provision model according to each behavior feature after association weighting; The information providing module is used to determine the information providing mode of each target information according to the target recommendation degree of each target information, and provide at least part of the target information to the user in the information providing mode.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.
11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.
