Method and apparatus for training network model for interactive behavior prediction

Abstract

The present disclosure provides a method and device for training a network model for interaction behavior prediction, wherein the method comprises: obtaining a sample behavior content sequence and a prompt text corresponding to the sample behavior content sequence; wherein a prompt result in the prompt text and a prediction result output by a network model to be trained are used to represent an interaction behavior prediction result corresponding to the sample behavior content sequence; inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence; and training the network model to be trained based on true value data corresponding to the prediction result and the prediction result determined when the prompt text is obtained, so as to perform interaction behavior prediction based on the trained network model.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of Internet, in particular, to a method and device for training a network model for interaction behavior prediction. BACKGROUND

[0002] With the development of Internet technology, the content contained on the network is also increasing. At present, the change of the content on the network can be predicted by a neural network model, and more content meeting the browsing demand of the user can be pushed based on the prediction result, thereby saving the search time of the user and improving the efficiency.

[0003] In the related art, the neural network model can be trained online, however, this training method requires a large amount of computing resources and has high computing delay, and the performance of the trained neural network model is poor, which is difficult to meet the prediction demand. SUMMARY

[0004] The present disclosure provides at least a method and device for training a network model for interaction behavior prediction.

[0005] In a first aspect, the present disclosure provides a method for training a network model for interaction behavior prediction, the method comprising: obtaining a sample behavior content sequence and a prompt text corresponding to the sample behavior content sequence; wherein a prompt result in the prompt text and a prediction result output by a to-be-trained network model are used to represent an interaction behavior prediction result corresponding to the sample behavior content sequence; inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the to-be-trained network model to obtain a prediction result corresponding to the sample behavior content sequence; training the to-be-trained network model based on true value data corresponding to the prediction result and the prediction result determined when obtaining the prompt text, to perform interaction behavior prediction based on the trained network model.

[0006] In an optional implementation, the prompt text corresponding to the sample behavior content sequence is determined by: obtaining a prompt text template matching an interaction behavior category predicted by the to-be-trained network model; obtaining interaction behavior annotation information corresponding to each sample behavior content sequence; wherein the interaction behavior annotation information comprises a prompt result and true value data corresponding to a to-be-predicted result; and determining the prompt text corresponding to each sample behavior content sequence based on the prompt text template and the prompt result in the interaction behavior annotation information corresponding to the sample behavior content sequence.

[0007] In an optional implementation, the inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence comprises: inputting each behavior content in the sample behavior content sequence into a multi-modal feature extraction model in the network model to be trained to obtain a first sequence feature corresponding to each behavior content; inputting the prompt text corresponding to the sample behavior content sequence into the multi-modal feature extraction model to obtain a first text feature; and determining the prediction result corresponding to the sample behavior content sequence based on the first sequence feature and the first text feature.

[0008] In an optional implementation, the network model to be trained comprises a content sequence model and a behavior prediction model; and the determining the prediction result corresponding to the sample behavior content sequence based on the first sequence feature and the first text feature comprises: inputting the first sequence feature into the content sequence model to obtain a deep sequence feature corresponding to each behavior content in the sample behavior content sequence; and inputting the first text feature and the deep sequence feature into the behavior prediction model to obtain the prediction result.

[0009] In an optional implementation, the sample behavior content sequence comprises an initial behavior content sequence and an enhanced behavior content sequence; and the method further comprises determining the enhanced behavior content sequence and a prompt text corresponding to the enhanced behavior content sequence according to the following method: performing enhancement processing on the initial behavior content sequence to obtain at least one enhanced behavior content sequence; and determining the prompt text corresponding to the enhanced behavior content sequence based on annotation information corresponding to the enhanced behavior content sequence.

[0010] In an optional implementation, the inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence comprises: inputting the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the enhanced behavior content sequence and a prediction enhancement processing method corresponding to the enhanced behavior content sequence; and the training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained and the prediction result comprises: training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained, the prediction result, the prediction enhancement processing method, and a real enhancement processing method corresponding to the enhanced behavior content sequence.

[0011] In an alternative implementation, the interaction behavior category includes any one of a sequence category analysis task, a sequence quality analysis task, a sequence interest change analysis task, and a sequence click prediction task.

[0012] In a second aspect, the embodiments of the present disclosure further provide a content pushing method, which comprises: obtaining a behavior content sequence to be predicted; inputting the behavior content sequence to be predicted into a target network model trained based on the training method of the network model for interaction behavior prediction according to the first aspect or any one of the possible implementation manners of the first aspect, to determine an interaction behavior prediction result corresponding to the behavior content sequence to be predicted; and pushing content based on the interaction behavior prediction result.

[0013] In a third aspect, the embodiments of the present disclosure further provide a training device of a network model for interaction behavior prediction, which comprises: a first obtaining unit configured to obtain a sample behavior content sequence and a prompt text corresponding to the sample behavior content sequence; wherein a prompt result in the prompt text and a prediction result output by a network model to be trained are used to represent an interaction behavior prediction result corresponding to the sample behavior content sequence; a prediction unit configured to input the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained, to obtain the prediction result corresponding to the sample behavior content sequence; and a training unit configured to train the network model to be trained based on true value data corresponding to the prediction result and the prediction result determined when the prompt text is obtained, to perform interaction behavior prediction based on the trained network model.

[0014] In a fourth aspect, the embodiments of the present disclosure further provide a content pushing device, which comprises: a second obtaining unit configured to obtain a behavior content sequence to be predicted; a determination unit configured to input the behavior content sequence to be predicted into a target network model trained based on the training method of the network model for interaction behavior prediction according to the first aspect or any one of the possible implementation manners of the first aspect, to determine an interaction behavior prediction result corresponding to the behavior content sequence to be predicted; and a pushing unit configured to push content based on the interaction behavior prediction result.

[0015] In a fifth aspect, the embodiments of the present disclosure further provide a computer device, which comprises a processor, a memory and a bus, the memory stores machine readable instructions executable by the processor, when the computer device is running, the processor and the memory communicate through the bus, the machine readable instructions are executed by the processor to execute the steps in the first aspect or any one of the possible implementation manners of the first aspect, or execute the steps in the second aspect.

[0016] In a sixth aspect, the present disclosure also provides a computer-readable storage medium, which stores a computer program. When the computer program is run by a processor, the steps of the first aspect or any possible implementation manner of the first aspect, or the steps of the second aspect are performed.

[0017] In the above embodiments, the sample behavior content sequence corresponding prompt text can be obtained, and the prompt text and the sample behavior content sequence are input into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence. Since the prompt result in the prompt text and the prediction result output by the network model to be trained are used to represent the interaction behavior prediction result corresponding to the sample behavior content sequence, and the prompt text is associated with the prediction task of the network model to be trained, the association between the sample behavior content sequence and the prediction task can be established based on the prompt text, so that the accuracy of the prediction result corresponding to the sample behavior content sequence obtained by the network model to be trained can be improved, and the training time of the network model to be trained is shortened, and the training efficiency of the network model to be trained is improved.

[0018] Further, the network model to be trained can be trained based on the true value data corresponding to the prediction result and the prediction result to obtain a trained network model, so that the training difficulty of the network model to be trained is reduced by offline training, and the training resources are saved. Then, the interaction behavior prediction can be performed based on the trained network model, so that the content pushing can be performed based on the interaction behavior prediction result, and the user experience is improved.

[0019] In order to make the above objectives, features and advantages of the present disclosure more apparent, the following will describe preferred embodiments in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following will briefly introduce the drawings needed to be used in the embodiments. The drawings are incorporated into the specification and form a part of the specification, which show the embodiments consistent with the present disclosure, and are used to explain the technical solutions of the present disclosure together with the specification. It should be understood that the following drawings only show some embodiments of the present disclosure, and therefore should not be considered as a limitation to the scope, and other related drawings can also be obtained by those skilled in the art without paying any creative labor.

[0021] Figure 1 A flowchart of a training method of a network model for interaction behavior prediction is shown.

[0022] Figure 2Fig. 1 shows a schematic diagram of determining a prediction result corresponding to a sample behavior content sequence based on first sequence features and first text features in a case where a network model to be trained includes a content sequence model and a behavior prediction model according to an embodiment of the present disclosure;

[0023] Figure 3 Fig. 2 shows a schematic diagram of a training process of a network model to be trained according to an embodiment of the present disclosure;

[0024] Figure 4 Fig. 3 shows a flowchart of a content pushing method according to an embodiment of the present disclosure;

[0025] Figure 5 Fig. 4 shows a schematic diagram of a training device of a network model for interactive behavior prediction according to an embodiment of the present disclosure;

[0026] Figure 6 Fig. 5 shows a schematic diagram of a content pushing device according to an embodiment of the present disclosure;

[0027] Figure 7 Fig. 6 shows a schematic diagram of a computer device according to an embodiment of the present disclosure;

[0028] Figure 8 Fig. 7 shows a schematic diagram of another computer device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] In order to make the objects, technical solutions and advantages of the embodiments of the present disclosure clearer, the following will be combined with the accompanying drawings for the embodiments of the present disclosure to clearly and completely describe the technical solutions of the embodiments of the present disclosure. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments of the present disclosure. The components of the embodiments of the present disclosure described and shown in the accompanying drawings can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present disclosure provided in the accompanying drawings is not intended to limit the scope of the claimed present disclosure, but only represents selected embodiments of the present disclosure. Based on the embodiments of the present disclosure, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present disclosure.

[0030] It should be noted that: similar reference numerals and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0031] The term "and / or", merely describes an associated relationship, which means that there can be three relationships, for example, A and / or B, which can represent the following three cases: A exists alone, A and B exist together, and B exists alone. In addition, the term "at least one" herein means any one of the plurality or any combination of at least two of the plurality, for example, including at least one of A, B, and C, which means including any one or more elements selected from the set consisting of A, B, and C.

[0032] It can be understood that before using the technical solutions disclosed in the embodiments of the present disclosure, the type, use range, use scenario, etc. of the personal information involved in the present disclosure should be informed to the user and the authorization of the user should be obtained in accordance with relevant laws and regulations.

[0033] It has been found through research that with the development of Internet technology, the content contained on the network is also increasing. Currently, the changes in the content on the network can be predicted by a neural network model, and more content that meets the browsing needs of the user can be pushed based on the prediction results, thereby saving the search time of the user and improving the efficiency.

[0034] In the related art, the neural network model can be trained online, however, this training method requires a large amount of computing resources and has high computing delay, and the performance of the trained neural network model is also poor, thereby making it difficult to meet the prediction requirements.

[0035] The inventors have also found that in the related art, the neural network model can also be trained in an offline manner, however, due to the large difference between the training samples of the neural network model and the training tasks corresponding to the neural network model, the training difficulty is high, thereby affecting the training efficiency of the neural network model.

[0036] Based on the above research, the present disclosure provides a method and device for training a network model for interactive behavior prediction. As described above, the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence can be obtained first, and then the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence are input into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence. Then, the network model to be trained is trained based on the true value data corresponding to the prediction result and the prediction result, and then the interactive behavior prediction can be performed based on the trained network model.

[0037] In the above embodiment, the sample behavior content sequence corresponding prompt text can be obtained, and the prompt text and the sample behavior content sequence can be input into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence. Since the prompt result in the prompt text and the prediction result output by the network model to be trained are used to represent the interaction behavior prediction result corresponding to the sample behavior content sequence, and the prompt text is associated with the prediction task of the network model to be trained, the association between the sample behavior content sequence and the prediction task can be established based on the prompt text, thereby improving the accuracy of the prediction result corresponding to the sample behavior content sequence obtained by the network model to be trained, shortening the training time of the network model to be trained, and improving the training efficiency of the network model to be trained.

[0038] Further, the network model to be trained can be trained based on the true value data corresponding to the prediction result and the prediction result, to obtain a trained network model, thereby reducing the training difficulty of the network model to be trained in an offline training manner, saving training resources. Then, the interaction behavior prediction can be performed based on the trained network model, so that the content pushing can be performed based on the interaction behavior prediction result, and the user experience is improved.

[0039] To facilitate the understanding of the present embodiment, first, a training method of a network model for interaction behavior prediction disclosed by the present embodiment is introduced in detail. The execution subject of the training method of the network model for interaction behavior prediction provided by the present embodiment is generally a computer device with certain computing power.

[0040] Referring to Figure 1 FIG. 1 shows a flowchart of a training method of a network model for interaction behavior prediction provided by the present embodiment. The method includes steps S101-S105, wherein:

[0041] S101: obtaining a sample behavior content sequence and a prompt text corresponding to the sample behavior content sequence; wherein the prompt result in the prompt text and the prediction result output by the network model to be trained are used to represent the interaction behavior prediction result corresponding to the sample behavior content sequence.

[0042] In the embodiments of the present disclosure, the sample behavior content sequence can be a post sequence, a click sequence or the like, and the sample behavior content sequence can include a plurality of sub-contents (i.e., behavior contents). The type of the sub-contents can be a video type, a text and image type, a text type or the like. The number of the sub-contents included in the sample behavior content sequence can be 10 or 20, and the present disclosure does not limit the number of the sub-contents included in the sample behavior content sequence. In addition, the number of the obtained sample behavior content sequences can be multiple, and the number of the sub-contents included in each sample behavior content sequence is the same. In the case that the number of the sub-contents included in the sample behavior content sequence is small, the empty content can be used for supplement to make the number of the sub-contents included in each sample behavior content sequence the same.

[0043] In the embodiments of the present disclosure, after obtaining the sample behavior content sequence, the prediction task of the network model to be trained based on the sample behavior content sequence can be determined. The prediction task of the network model to be trained can be a task of predicting the interaction behavior category. For example, in the case that the sample behavior content sequence is a post sequence, the prediction task (i.e., the interaction behavior category) of the network model to be trained can be a post sequence category analysis (i.e., sequence category analysis), a post sequence quality analysis (i.e., sequence quality analysis), a post sequence interest change analysis (i.e., sequence interest change analysis) or the like. In the case that the sample behavior content sequence is a click sequence, the prediction task (i.e., the interaction behavior category) of the network model to be trained can be a sequence click analysis or the like.

[0044] In the embodiments of the present disclosure, after determining the prediction task corresponding to the network model to be trained, the prompt text corresponding to the sample behavior content sequence can be obtained. Specifically, the interaction behavior annotation information established based on the sample behavior content sequence can be obtained, and the interaction behavior annotation information can include at least one annotation result. Each annotation result is associated with the specific content of each sub-content in the sample behavior content sequence and the prediction task of the network model to be trained. Then, the part of the annotation results in the at least one annotation result can be determined as the prompt text corresponding to the sample behavior content sequence.

[0045] Here, the part of the annotation results included in the prompt text is the prompt result in the prompt text, and the number of the prompt results included in the prompt text can be one or multiple.

[0046] For example, in a case where the prediction task of the network model to be trained is a sequence category analysis task, and each sub-content of the sample behavior content sequence is associated with football, the prompt result in the prompt text corresponding to the sample behavior content sequence can be one, for example, the prompt result can be "sequence category is XX football team", or the prompt result in the prompt text can also be two, for example, the prompt result can be "sequence category level one is sports, and sequence category level two is football".

[0047] S103: inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence.

[0048] Here, the network model to be trained can be understood as a network model for processing different modal data, for example, the network model to be trained can be a double-tower multi-modal network model.

[0049] In this case, the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence can be input into the network model to be trained for processing and prediction of different modal data to obtain a prediction result corresponding to the sample behavior content sequence.

[0050] S105: training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained and the prediction result, to perform interactive behavior prediction based on the trained network model.

[0051] As described above, in a case where the prompt text corresponding to the sample behavior content sequence is obtained, the interactive behavior annotation information established for the sample behavior content sequence can be obtained first, and the prompt text corresponding to the sample behavior content sequence can be determined based on part of the annotation results in the interactive behavior annotation information. At this time, the true value data corresponding to the prediction result can also be determined based on the annotation results other than the part of the annotation results in the at least one annotation result included in the interactive behavior annotation information.

[0052] For example, in a case where the interactive behavior annotation information contains 3 annotation results, the 3 annotation results are annotation result 1 "football", annotation result 2 "sports", and annotation result 3 "ball", annotation result 2 "sports" and annotation result 3 "ball" can be determined as the prompt text corresponding to the sample behavior content sequence (i.e., the prompt text can be "sequence category level one is sports, and level two is ball"), and annotation result 1 "football" can be determined as the true value data corresponding to the prediction result, i.e., the true value data is "sequence category is football".

[0053] In the embodiments of the present disclosure, the to-be-trained network model can be trained based on a prediction result output by the to-be-trained network model and true value data corresponding to the prediction result, to obtain a trained network model. Then, the real behavior content sequence can be input into the trained network model, so that the trained network model performs interactive behavior prediction on the real behavior content sequence.

[0054] As can be known from the above description, the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence can be acquired first, and then the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence are input into the to-be-trained network model, to obtain a prediction result corresponding to the sample behavior content sequence. Then, the to-be-trained network model is trained based on true value data corresponding to the prediction result and the prediction result, and then, the trained network model can be used for interactive behavior prediction.

[0055] In the above embodiments, the prompt text corresponding to the sample behavior content sequence can be acquired, and the prompt text and the sample behavior content sequence are input into the to-be-trained network model, to obtain a prediction result corresponding to the sample behavior content sequence. Since the prompt result in the prompt text and the prediction result output by the to-be-trained network model are used to represent the interactive behavior prediction result corresponding to the sample behavior content sequence, and the prompt text is associated with the prediction task of the to-be-trained network model, the association relationship between the sample behavior content sequence and the prediction task can be established based on the prompt text, so that the accuracy of the prediction result corresponding to the sample behavior content sequence obtained by the to-be-trained network model can be improved, and the training time of the to-be-trained network model is shortened, and the training efficiency of the to-be-trained network model is improved.

[0056] Further, the to-be-trained network model can be trained based on true value data corresponding to the prediction result and the prediction result, to obtain a trained network model, so that the training difficulty of the to-be-trained network model is reduced in an offline training manner, and training resources are saved. Then, the trained network model can be used for interactive behavior prediction, so that content pushing can be performed based on the interactive behavior prediction result, and user experience is improved.

[0057] In an optional embodiment, the prompt text corresponding to the sample behavior content sequence can be determined in the following manner:

[0058] First, a prompt text template matching an interactive behavior category predicted by the to-be-trained network model is acquired.

[0059] In the embodiments of the present disclosure, the template form of the prompt text template can be a fill-in-the-blank template, at this time, the prompt result in the prompt text can be the fill-in content, or the prompt text template can also be a prefix template, at this time, the prefix template indicates that the prompt result in the prompt text is located at the front of the prompt text, and the present disclosure does not make specific limitations on the template form of the prompt text template, and the actual needs are used as the criterion.

[0060] In the embodiments of the present disclosure, the template form of the prompt text template can be determined first, and then the prompt text template matched with the predicted interactive behavior category of the network model to be trained is obtained, for example, in the case of sequence category analysis of the interactive behavior category, the prompt text template matched with the sequence category analysis is a sequence category template, for example, the sequence category template can be “the first level of the sequence category is [y], and the second level is [x]”. Or, in the case of sequence quality analysis of the interactive behavior category, the prompt text template matched with the sequence quality analysis is a sequence quality template, for example, the sequence quality template can be “[x] sub-contents are ordinary quality, and the next [y] sub-contents are high-quality quality”, and the like, which will not be listed one by one here.

[0061] In the embodiments of the present disclosure, since the number of sample behavior content sequences used for training the network model to be trained can be multiple, for any sample behavior content sequence, the interactive behavior annotation information corresponding to the sample behavior content sequence is obtained; wherein the interactive behavior annotation information includes a prompt result and true value data corresponding to a to-be-predicted result.

[0062] In the embodiments of the present disclosure, the interactive behavior annotation information corresponding to the sample behavior content sequence is associated with the interactive behavior category of the network model to be trained, for example, in the case of sequence category analysis of the interactive behavior category, the interactive behavior annotation information corresponding to the sample behavior content sequence can be category information, for example, the interactive behavior annotation information can be “ball”, “sports”, “soccer”, and the like. Or, in the case of sequence quality analysis of the interactive behavior category, the interactive behavior annotation information corresponding to the sample behavior content sequence can be quality information, for example, the interactive behavior annotation information can be that X sub-contents are high-quality quality, and / or Y sub-contents are ordinary quality, and the like, which will not be listed one by one here.

[0063] Then, based on the prompt text template and the prompt result in the interactive behavior annotation information corresponding to the sample behavior content sequence, the prompt text corresponding to the sample behavior content sequence is determined.

[0064] In the embodiments of the present disclosure, the prompt result in the interactive behavior annotation information can be put into the corresponding position in the prompt text template to obtain the prompt text corresponding to the sample behavior content sequence.

[0065] For example, in the case that the prompt text template is a sequence category template: "sequence category level one is [y], level two is [x]", and the prompt result in the interactive behavior annotation information is "sports" and "ball", then the determined prompt text corresponding to the sample behavior content sequence is: "sequence category level one is sports, level two is ball". Or, in the case that the prompt text template is a sequence category template: "sequence category level two is [x]", and the prompt result in the interactive behavior annotation information is "ball", then the determined prompt text corresponding to the sample behavior content sequence is: "sequence category level two is ball".

[0066] In the above embodiment, the prompt text template matched with the interactive behavior category predicted by the network model to be trained can be obtained, and the interactive behavior annotation information corresponding to the sample behavior content sequence is obtained, so that the prompt text corresponding to the sample behavior content sequence is determined based on the prompt text template and the prompt result in the interactive behavior annotation information. Therefore, the relevance of the prompt text and the interactive behavior category predicted by the network model to be trained can be improved, and the accuracy of the prompt result in the prompt text can be improved, so that the prediction result output by the network model to be trained is more matched with the interactive behavior category predicted by the network model to be trained, and the accuracy is higher.

[0067] In an optional embodiment, for the above S103: inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence, the following steps are included:

[0068] Step S21: inputting each behavior content in the sample behavior content sequence into a multi-modal feature extraction model in the network model to be trained to obtain a first sequence feature corresponding to each behavior content, respectively;

[0069] Step S22: inputting the prompt text corresponding to the sample behavior content sequence into the multi-modal feature extraction model to obtain a first text feature;

[0070] Step S23: determining the prediction result corresponding to the sample behavior content sequence based on the first sequence feature and the first text feature.

[0071] In the embodiments of the present disclosure, the network model to be trained can include a multi-modal extraction model, and the multi-modal extraction model can be used to extract initial features corresponding to multi-modal data. At this time, the multi-modal extraction model can include at least one sub-network model, and each sub-network model can be used to extract features of at least one modal data. At this time, the network structures of each sub-network model can be the same or different.

[0072] In this case, the sample behavior content sequence and the corresponding prompt text can be input into the multimodal feature extraction model to obtain the first sequence features corresponding to each sub-content (i.e. behavior content) in the sample behavior content sequence and the first text features corresponding to the prompt text.

[0073] In this embodiment of the disclosure, after obtaining the first sequence features and the first text features, the prediction result corresponding to the sample behavior content sequence can be determined based on the first sequence features and the first text features.

[0074] In this case, the network model to be trained may also include a content sequence model and a behavior prediction model. In this case, the first sequence features can be input into the content sequence model to obtain the deep sequence features corresponding to each behavior content of the sample behavior content sequence; and the first text features and the above-mentioned deep sequence features can be input into the behavior prediction model to obtain the prediction result.

[0075] For example, such as Figure 2 As shown, assuming that the number of behaviors contained in the sample behavior content sequence is N, then after obtaining the first sequence feature corresponding to each behavior content (that is, the first sequence feature 1 corresponding to behavior content 1, the first sequence feature 2 corresponding to behavior content 2, ..., the first sequence feature N corresponding to behavior content N), these N first sequence features can be input into the content sequence model to obtain the deep sequence features corresponding to each behavior content of the sample behavior content sequence, that is, the deep sequence feature 1 corresponding to behavior content 1, the deep sequence feature 2 corresponding to behavior content 2, ..., the deep sequence feature N corresponding to behavior content N, etc.

[0076] After that, as Figure 2 As shown, the first text feature corresponding to the prompt text and the deep sequence feature obtained above can be input into the behavior prediction model. The behavior prediction model performs interactive processing on the first text feature and the deep sequence feature (e.g., concatenation, mean processing, pooling, etc.) to obtain the prediction result.

[0077] In the above embodiments, the first sequence features corresponding to each behavior content in the sample behavior content sequence and the first text features of the prompt text corresponding to the sample behavior content sequence can be extracted by the multi-modal extraction model in the network model to be trained, so that the feature extraction of the multi-modal data can be realized, thereby improving the accuracy and efficiency of obtaining the first sequence features and the first text features. Then, the first sequence features can be input into the content sequence model to obtain deep sequence features corresponding to each behavior content in the sample behavior content sequence, and the first text features and the deep sequence features can be input into the behavior prediction model to obtain the prediction result, so that the accuracy of the obtained prediction result can be improved by combining the first text features and the deep sequence features, thereby improving the training efficiency and training accuracy of the network model to be trained.

[0078] In an optional embodiment, the sample behavior content sequence includes an initial behavior content sequence and an enhanced behavior content sequence. Based on this, the disclosure embodiments can determine the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence according to the following process:

[0079] The initial behavior content sequence is enhanced to obtain at least one enhanced behavior content sequence; and based on the annotation information corresponding to the enhanced behavior content sequence, the prompt text corresponding to the enhanced behavior content sequence is determined.

[0080] In the embodiments of the disclosure, real interaction data of a user can be obtained, and the initial behavior content sequence is determined based on the real interaction data. For example, the initial behavior content sequence can be 10 sequences of posts of the user, or 10 sequences of continuous clicks of the user, etc. Here, the process of obtaining the real interaction data of the user can refer to the description in paragraph 4 of the specific embodiments.

[0081] Here, the initial behavior content sequence can include one or more behavior contents (i.e., sub-contents).

[0082] In the embodiments of the disclosure, after obtaining the initial behavior content sequence, the initial behavior content sequence can be enhanced according to at least one preset enhancement processing method to obtain at least one enhanced behavior content sequence. For example, the above-mentioned preset enhancement processing method can be a repeated content sequence method, an adjusted content sequence order method, a content sequence replacement method (for example, one or more behavior contents in the initial behavior content sequence can be replaced by empty data), etc.

[0083] In the embodiments of the disclosure, the annotation information corresponding to each enhanced behavior content sequence can be obtained, and the prompt text corresponding to the enhanced behavior content sequence is determined based on the annotation information.

[0084] Specifically, the annotation information corresponding to the enhanced behavior content sequence can be determined according to the preset enhancement processing method corresponding to the enhanced behavior content sequence and the annotation information corresponding to the initial behavior content sequence.

[0085] For example, in a case where the preset enhancement processing method corresponding to the enhanced behavior content sequence is the repeated content sequence method, the annotation information corresponding to the initial behavior content sequence can be determined as the annotation information corresponding to the enhanced behavior content sequence.

[0086] Alternatively, in a case where the preset enhancement processing method corresponding to the enhanced behavior content sequence is the adjusted content sequence order method, the annotation information corresponding to the enhanced behavior content sequence can be determined based on the interaction behavior category predicted by the network model to be trained and the annotation information corresponding to the initial behavior content sequence. For example, in a case where the interaction behavior category predicted by the network model to be trained is sequence category analysis, the annotation information corresponding to the initial behavior content sequence can be determined as the annotation information corresponding to the enhanced behavior content sequence; in a case where the interaction behavior category predicted by the network model to be trained is sequence quality analysis, the annotation information corresponding to the enhanced behavior content sequence can be re-determined based on the annotation information corresponding to the initial behavior content sequence. Here, the above cases are not listed one by one.

[0087] In the above embodiments, the initial behavior content sequence can be enhanced to obtain at least one enhanced behavior content sequence, so as to improve the utilization rate of the initial behavior content sequence and save the acquisition cost of the sample behavior content sequence. Then, the initial behavior content sequence and the enhanced behavior content sequence can be input into the network model to be trained for processing, so as to expand the data amount of the training sample for training the network model to be trained and improve the training precision and efficiency of the network model to be trained.

[0088] In an optional embodiment, in a case where the sample behavior content sequence includes the initial behavior content sequence and the enhanced behavior content sequence, the above S103: inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence, specifically includes the following process:

[0089] inputting the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence into the network model to be trained to obtain the prediction result corresponding to the enhanced behavior content sequence and the prediction enhancement processing method corresponding to the enhanced behavior content sequence.

[0090] In the embodiments of the present disclosure, the enhanced behavior content sequence can also include at least one sub-content (i.e., behavior content), and when the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence are input into the network model to be trained, the prediction enhancement processing method corresponding to each behavior content in the enhanced behavior content sequence can be obtained.

[0091] For example, as shown in Figure 2 , assuming that, in the N behavior contents included in the sample behavior content sequence, the behavior content M is the behavior content included in the enhanced behavior content sequence, the prediction enhancement processing method corresponding to the behavior content M can be obtained through the content sequence model, i.e., the prediction enhancement processing method as shown in Figure 2 .

[0092] At this time, for the above S105: training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained and the prediction result, the specific process includes the following steps:

[0093] training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained, the prediction result, the prediction enhancement processing method, and the real enhancement processing method corresponding to the enhanced behavior content sequence.

[0094] In the embodiments of the present disclosure, after the initial behavior content sequence is enhanced, the real enhancement processing method corresponding to the enhanced behavior content sequence can be determined based on the correspondence between the enhanced behavior content sequence and the initial behavior content sequence.

[0095] In the above embodiments, the network model to be trained can be trained based on the true value data corresponding to the prediction result, the prediction result, the prediction enhancement processing method, and the real enhancement processing method corresponding to the prediction enhancement processing method, so that the network model to be trained can not only obtain the prediction result corresponding to the sample behavior content sequence, but also obtain the enhancement processing method corresponding to the enhanced behavior content sequence, thereby improving the generalization ability of the network model to be trained, and further expanding the application range of the network model to be trained and improving the performance of the network model to be trained.

[0096] The detailed process of the training method of the network model for interaction behavior prediction provided by the present disclosure will be introduced below in combination with the training flowchart of the network model to be trained as shown in Figure 3 .

[0097] As shown in Figure 3As shown, first, after determining the prediction task of the network model to be trained (i.e., the interaction behavior category predicted by the network model to be trained), an initial behavior content sequence can be obtained, and an enhanced behavior content sequence can be determined based on the initial behavior content sequence. At this time, the initial behavior content sequence and the enhanced behavior content sequence after merging are the above-mentioned sample behavior content sequence.

[0098] At this time, a prompt text template matching the interaction behavior category predicted by the network model to be trained can also be obtained, as well as the interaction behavior annotation information corresponding to the sample behavior content sequence, and the prompt text corresponding to the sample behavior content sequence can be determined based on the prompt result in the prompt text template and the interaction behavior annotation information corresponding to the sample behavior content sequence.

[0099] Next, each behavior content in the sample behavior content sequence can be input into the multi-modal feature extraction model in the network model to be trained to obtain the first sequence feature corresponding to each behavior content, and the prompt text corresponding to the sample behavior content sequence can be input into the multi-modal feature extraction model to obtain the first text feature.

[0100] Then, the prediction result corresponding to the sample behavior content sequence can be determined based on the obtained first sequence feature and first text feature, and the network model to be trained can be trained based on the prediction result and the true value data corresponding to the prediction result to obtain the trained network model.

[0101] Referring to Figure 4 As shown, a flowchart of a method for training a network model for interaction behavior prediction is provided in the embodiments of the present disclosure, and the method comprises steps S401-S405, wherein:

[0102] Step S401: obtaining a behavior content sequence to be predicted;

[0103] Step S403: inputting the behavior content sequence to be predicted into a target network model trained based on the above-mentioned any one method for training a network model for interaction behavior prediction to determine an interaction behavior prediction result corresponding to the behavior content sequence to be predicted;

[0104] Step S405: performing content pushing based on the interaction behavior prediction result.

[0105] In the embodiments of the present disclosure, the behavior content sequence to be predicted can be understood as a sequence of interaction behaviors of a user, wherein the behavior content sequence to be predicted can contain at least one behavior content (i.e., sub-content), and the content type of the behavior content can be a video type, a text and image type, or a text type, etc. The method for obtaining the behavior content sequence to be predicted is described in the 4th paragraph of the specific embodiments.

[0106] In the embodiments of the present disclosure, the behavior content sequence to be predicted can be input into the target network model, the interaction behavior prediction result corresponding to the behavior content sequence to be predicted is determined, and content pushing is performed based on the interaction behavior prediction result. For example, in a case where it is determined that the interaction behavior prediction result is that the sequence category analysis result is "football", then sports content pushing or football content pushing can be performed for the user based on the interaction behavior prediction result. Or, in a case where it is determined that the interaction behavior prediction result is that the sequence quality analysis result is that the last 4 behavior contents are high-quality contents, then contents associated with the high-quality contents can be pushed to the user, or the high-quality contents of the user can be pushed to other users, and the like, which are not listed one by one here.

[0107] In the above-mentioned embodiments, the behavior content sequence to be predicted can be obtained, and the interaction behavior prediction result corresponding to the behavior content to be predicted is determined based on the trained target network model, so that content pushing can be performed based on the interaction behavior prediction result, thereby improving the accuracy and efficiency of content pushing, and further improving the browsing experience of the user.

[0108] Those skilled in the art can understand that in the above-mentioned method of the specific embodiments, the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process, and the specific execution order of each step should be determined by its function and possible internal logic.

[0109] Based on the same inventive concept, the embodiments of the present disclosure also provide a training device for a network model for interaction behavior prediction corresponding to the training method of the network model for interaction behavior prediction. Since the principle of solving problems in the device of the embodiments of the present disclosure is similar to the training method of the network model for interaction behavior prediction of the above-mentioned embodiments of the present disclosure, the implementation of the device can be referred to the implementation of the method, and the repeated parts will not be described here.

[0110] Referring to Figure 5 Fig. 1 shows a schematic diagram of a training device for a network model for interaction behavior prediction provided by the embodiments of the present disclosure, the device comprises: a first acquisition unit 51, a prediction unit 52, and a training unit 53; wherein,

[0111] The first acquisition unit 51 is configured to acquire a sample behavior content sequence and a prompt text corresponding to the sample behavior content sequence; wherein, the prompt result in the prompt text and the prediction result output by the network model to be trained are used to represent the interaction behavior prediction result corresponding to the sample behavior content sequence;

[0112] The prediction unit 52 is configured to input the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence.

[0113] The training unit 53 is configured to train the network model to be trained based on the true value data corresponding to the prediction result and the prediction result determined when the prompt text is obtained, and perform interactive behavior prediction based on the trained network model.

[0114] In the above embodiment, the prompt text corresponding to the sample behavior content sequence is obtained, and the prompt text and the sample behavior content sequence are input into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence. Since the prompt result in the prompt text and the prediction result output by the network model to be trained are used to represent the interactive behavior prediction result corresponding to the sample behavior content sequence, the prompt text is associated with the prediction task of the network model to be trained, so that the association relationship between the sample behavior content sequence and the prediction task can be established based on the prompt text, thereby improving the accuracy of the prediction result corresponding to the sample behavior content sequence obtained by the network model to be trained, shortening the training time of the network model to be trained, and improving the training efficiency of the network model to be trained.

[0115] Further, the network model to be trained can be trained based on the true value data corresponding to the prediction result and the prediction result to obtain a trained network model, so that the training difficulty of the network model to be trained is reduced by offline training, and training resources are saved. Then, the interactive behavior prediction can be performed based on the trained network model, so that content pushing can be performed based on the interactive behavior prediction result, and user experience is improved.

[0116] In a possible implementation, the apparatus is further configured to: determine the prompt text corresponding to the sample behavior content sequence by: obtaining a prompt text template matching the interactive behavior category predicted by the network model to be trained; obtaining, for any sample behavior content sequence, interactive behavior annotation information corresponding to the sample behavior content sequence; wherein the interactive behavior annotation information includes a prompt result and true value data corresponding to the prediction result; and determining the prompt text corresponding to the sample behavior content sequence based on the prompt text template and the prompt result in the interactive behavior annotation information corresponding to the sample behavior content sequence.

[0117] In a possible implementation, the prediction unit 52 is further configured to: input each behavior content in the sample behavior content sequence into a multi-modal feature extraction model in the network model to be trained, to obtain a first sequence feature corresponding to each behavior content; input the prompt text corresponding to the sample behavior content sequence into the multi-modal feature extraction model, to obtain a first text feature; and determine the prediction result corresponding to the sample behavior content sequence based on the first sequence feature and the first text feature.

[0118] In a possible implementation, the prediction unit 52 is further configured to: in a case where the network model to be trained includes a content sequence model and a behavior prediction model, input the first sequence feature into the content sequence model, to obtain a deep sequence feature corresponding to each behavior content in the sample behavior content sequence; and input the first text feature and the deep sequence feature into the behavior prediction model, to obtain the prediction result.

[0119] In a possible implementation, the apparatus is further configured to: in a case where the sample behavior content sequence includes an initial behavior content sequence and an enhanced behavior content sequence, determine the enhanced behavior content sequence and a prompt text corresponding to the enhanced behavior content sequence according to the following method: performing enhancement processing on the initial behavior content sequence to obtain at least one enhanced behavior content sequence; and determining the prompt text corresponding to the enhanced behavior content sequence based on annotation information corresponding to the enhanced behavior content sequence.

[0120] In a possible implementation, the prediction unit 52 is further configured to: input the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence into the network model to be trained, to obtain a prediction result corresponding to the enhanced behavior content sequence and a prediction enhancement processing method corresponding to the enhanced behavior content sequence; and train the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained and the prediction result, including: training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained, the prediction result, the prediction enhancement processing method, and a real enhancement processing method corresponding to the enhanced behavior content sequence.

[0121] In a possible implementation, the interaction behavior category includes any one of a sequence category analysis task, a sequence quality analysis task, a sequence interest change analysis task, and a sequence click prediction task.

[0122] This disclosure also provides a content push device corresponding to the content push method. Since the principle of the device in this disclosure is similar to the content push method described above, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0123] Reference Figure 6 The diagram shown is a schematic representation of a content push device provided in an embodiment of this disclosure. The device includes: a second acquisition unit 61, a determination unit 62, and a push unit 63; wherein,

[0124] The second acquisition unit 61 is used to acquire the sequence of behavior content to be predicted;

[0125] The determining unit 62 is used to input the sequence of behavior content to be predicted into the target network model trained based on the training method for the network model for interactive behavior prediction described above, and to determine the interactive behavior prediction result corresponding to the sequence of behavior content to be predicted.

[0126] Push unit 63 is used to push content based on the interaction behavior prediction results.

[0127] In the above embodiments, a sequence of content to be predicted can be obtained, and the prediction result of the interactive behavior corresponding to the content to be predicted can be determined based on the trained target network model. Thus, content can be pushed based on the prediction result of the interactive behavior, thereby improving the accuracy and efficiency of content push and enhancing the user's browsing experience.

[0128] The processing flow of each module in the device and the interaction flow between each module can be referred to the relevant descriptions in the above method embodiments, and will not be detailed here.

[0129] Corresponding to Figure 1 The present disclosure also provides a computer device 700, such as a method for training a network model for predicting interactive behavior. Figure 7 The diagram shown is a structural schematic of a computer device 700 provided in an embodiment of this disclosure, including:

[0130] The computer device 700 comprises a processor 71, a memory 72, and a bus 73. The memory 72 stores execution instructions and includes main memory 721 and external memory 722. The main memory 721, also called internal memory, temporarily stores the processing data in the processor 71, as well as data exchanged with external memory such as a hard disk. The processor 71 exchanges data with the external memory 722 through the main memory 721. When the computer device 700 is running, the processor 71 communicates with the memory 72 through the bus 73, causing the processor 71 to execute the following instructions:

[0131] obtaining a sample behavior content sequence and prompt text corresponding to the sample behavior content sequence; wherein a prompt result in the prompt text and a prediction result output by the network model to be trained are used to represent an interaction behavior prediction result corresponding to the sample behavior content sequence;

[0132] inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence;

[0133] training the network model to be trained based on true value data corresponding to the prediction result and the prediction result determined when the prompt text is obtained, to perform interaction behavior prediction based on the trained network model.

[0134] corresponding to Figure 4 The embodiment of the present disclosure also provides another computer device 800, as shown in Figure 8 a structural schematic diagram of the computer device 800 provided by the embodiment of the present disclosure, which comprises:

[0135] a processor 81, a memory 82, and a bus 83; the memory 82 is used for storing execution instructions, and comprises an internal memory 821 and an external memory 822; the internal memory 821 is also referred to as an internal storage, and is used for temporarily storing operation data in the processor 81 and data exchanged with the external memory 822 such as a hard disk; the processor 81 exchanges data with the external memory 822 through the internal memory 821; when the computer device 800 is running, the processor 81 and the memory 82 communicate through the bus 83, so that the processor 81 executes the following instructions:

[0136] obtaining a behavior content sequence to be predicted;

[0137] inputting the behavior content sequence to be predicted into a target network model trained by the training method of the network model for performing interaction behavior prediction according to any one of the above embodiments to determine an interaction behavior prediction result corresponding to the behavior content sequence to be predicted;

[0138] performing content pushing based on the interaction behavior prediction result.

[0139] The embodiment of the present disclosure also provides a computer readable storage medium, which stores a computer program, and the computer program is run by a processor to execute the steps of the training method of the network model for performing interaction behavior prediction in the method embodiment or execute the steps of the content pushing method in the method embodiment. The storage medium can be a volatile or non-volatile computer readable storage medium.

[0140] The embodiment of the present disclosure further provides a computer program product carrying program codes, the program codes comprising instructions for performing the steps of the training method of the network model for interaction behavior prediction described in the above method embodiments, or performing the steps of the content pushing method described in the above method embodiments. For details, refer to the above method embodiments, which will not be repeated here.

[0141] The computer program product can be implemented by hardware, software or a combination thereof. In an optional embodiment, the computer program product is embodied as a computer storage medium. In another optional embodiment, the computer program product is embodied as a software product, such as a software development kit (SDK) and the like.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described system and device can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here. In several embodiments provided by the present disclosure, it should be understood that the disclosed system, device and method can be implemented by other ways. The above-described device embodiments are only schematic, for example, the division of the units is only a logical function division, and other division manners can be used in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some communication interfaces, devices or units, and can be electrical, mechanical or other forms.

[0143] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., they can be located in one place or distributed on multiple network units. According to actual needs, some or all of the units can be selected to achieve the purpose of the present embodiment.

[0144] In addition, each functional unit in each embodiment of the present disclosure can be integrated into one processing unit, or each unit can exist physically, or two or more units can be integrated into one unit.

[0145] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a nonvolatile computer readable storage medium executable by a processor. Based on this understanding, the technical solutions of the present disclosure essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0146] Finally, it should be noted that: the above-described embodiments are only specific embodiments of the present disclosure, used to illustrate the technical solutions of the present disclosure, and not to limit them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can make modifications or easily think of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed by the present disclosure, or make equivalent replacements to some of the technical features. The modifications, changes or replacements do not cause the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A method for training a network model for predicting interactive behavior, characterized in that, The method comprises: obtaining a sample behavior content sequence and prompt text corresponding to the sample behavior content sequence; wherein a prompt result in the prompt text and a prediction result output by a network model to be trained are used to represent an interaction behavior prediction result corresponding to the sample behavior content sequence; the prompt text is associated with a prediction task of the network model to be trained; inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence; training the network model to be trained based on true value data corresponding to the prediction result and the prediction result determined when the prompt text is obtained, so as to perform prediction of an interaction behavior category based on a trained network model, wherein the prompt text is determined based on part of the annotation results in the interaction behavior annotation information established based on the sample behavior content sequence; the true value data is determined based on annotation results other than the part of the annotation results in the annotation results included in the interaction behavior annotation information; the interaction behavior annotation information is associated with the interaction behavior category of the network model to be trained; wherein the sample behavior content sequence comprises at least one behavior content, and the content type of the behavior content comprises a video type, a picture-text type or a text type.

2. The method of claim 1, wherein, The prompt text corresponding to the sample behavior content sequence is determined in the following manner: obtaining a prompt text template matching the interaction behavior category predicted by the network model to be trained; for any sample behavior content sequence, obtaining interaction behavior annotation information corresponding to the sample behavior content sequence; wherein the interaction behavior annotation information comprises a prompt result and true value data corresponding to a prediction result to be predicted; determining the prompt text corresponding to the sample behavior content sequence based on the prompt text template and the prompt result in the interaction behavior annotation information corresponding to the sample behavior content sequence.

3. The method of claim 1, wherein, The inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence comprises: inputting each behavior content in the sample behavior content sequence into a multi-modal feature extraction model in the network model to be trained to obtain a first sequence feature corresponding to each behavior content; inputting the prompt text corresponding to the sample behavior content sequence into the multi-modal feature extraction model to obtain a first text feature; determining the prediction result corresponding to the sample behavior content sequence based on the first sequence feature and the first text feature.

4. The method of claim 3, wherein, The network model to be trained comprises a content sequence model and a behavior prediction model; The determining the prediction result corresponding to the sample behavior content sequence based on the first sequence feature and the first text feature comprises: inputting the first sequence feature into the content sequence model to obtain a deep sequence feature corresponding to each behavior content in the sample behavior content sequence; inputting the first text feature and the deep sequence feature into the behavior prediction model to obtain the prediction result.

5. The method of claim 1, wherein, The sample behavior content sequence includes an initial behavior content sequence and an enhanced behavior content sequence. The method further includes determining the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence according to the following method: performing enhancement processing on the initial behavior content sequence to obtain at least one enhanced behavior content sequence; and determining the prompt text corresponding to the enhanced behavior content sequence based on the annotation information corresponding to the enhanced behavior content sequence.

6. The method of claim 5, wherein, The inputting the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain the prediction result corresponding to the sample behavior content sequence includes: inputting the enhanced behavior content sequence and the prompt text corresponding to the enhanced behavior content sequence into the network model to be trained to obtain the prediction result corresponding to the enhanced behavior content sequence and a prediction enhancement processing method corresponding to the enhanced behavior content sequence; The training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained and the prediction result includes: training the network model to be trained based on the true value data corresponding to the prediction result determined when the prompt text is obtained, the prediction result, the prediction enhancement processing method, and a real enhancement processing method corresponding to the enhanced behavior content sequence.

7. The method of claim 1, wherein, The interaction behavior category includes any one of a sequence category analysis task, a sequence quality analysis task, a sequence interest change analysis task, and a sequence click prediction task.

8. A content push method characterized by, The method includes: obtaining a behavior content sequence to be predicted; inputting the behavior content sequence to be predicted into a target network model trained based on the training method for training the network model for interaction behavior prediction according to any one of claims 1 to 7 to determine an interaction behavior prediction result corresponding to the behavior content sequence to be predicted; and performing content pushing based on the interaction behavior prediction result.

9. A training apparatus for a network model used for predicting interactive behavior, characterized in that, The method includes: a first obtaining unit configured to obtain a sample behavior content sequence and a prompt text corresponding to the sample behavior content sequence; wherein a prompt result in the prompt text and a prediction result output by a network model to be trained are used to represent an interaction behavior prediction result corresponding to the sample behavior content sequence; and the prompt text is associated with a prediction task of the network model to be trained; a prediction unit configured to input the sample behavior content sequence and the prompt text corresponding to the sample behavior content sequence into the network model to be trained to obtain a prediction result corresponding to the sample behavior content sequence; a training unit configured to train the network model to be trained based on true value data corresponding to the prediction result determined when the prompt text is obtained and the prediction result, so as to perform prediction of an interaction behavior category based on the trained network model. The prompt text is determined based on part of the annotation results in the interaction behavior annotation information established based on the sample behavior content sequence; the true value data is determined based on annotation results other than the part of the annotation results in the interaction behavior annotation information; the interaction behavior annotation information is associated with an interaction behavior category of the network model to be trained. The sample behavior content sequence includes at least one behavior content, and a content type of the behavior content includes a video type, a picture-text type, or a text type.

10. A content pusher device, characterized by The method comprises the following steps: The second acquisition unit is configured to acquire a behavior content sequence to be predicted; The determination unit is configured to input the behavior content sequence to be predicted into a target network model trained based on the training method for interaction behavior prediction of the network model according to any one of claims 1 to 7, and determine an interaction behavior prediction result corresponding to the behavior content sequence to be predicted; The pushing unit is configured to perform content pushing based on the interaction behavior prediction result.

11. A computer device, comprising: The computer device comprises a processor, a memory, and a bus, the memory stores machine readable instructions executable by the processor, when the computer device is running, the processor and the memory communicate through the bus, the machine readable instructions are executed by the processor to perform the steps of the training method for the network model for interaction behavior prediction according to any one of claims 1 to 7, or perform the steps of the content pushing method according to claim 8. The computer readable storage medium stores a computer program, when the computer program is run by the processor, the steps of the training method for the network model for interaction behavior prediction according to any one of claims 1 to 7 are executed, or the steps of the content pushing method according to claim 8 are executed.

12. A computer-readable storage medium, characterized in that, ​

Images