Content recommendation and intention recognition model training method and device, equipment and medium

Abstract

The invention provides a content recommendation and intention recognition model training method and device, equipment and a medium, and relates to the technical field of artificial intelligence, in particular to the technical fields of intelligent recommendation, deep learning, large models and the like. The content recommendation method comprises the steps of determining a current intention of a target object; in response to determining that the current intention is an extension intention, obtaining associated content of entry content; the extension intention is used for indicating that the degree of satisfaction of the target object to the associated content is higher than the degree of satisfaction of non-associated content; and recommending the associated content to the target object.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, specifically to the technical fields of intelligent recommendation, deep learning, large models, etc., and in particular to a content recommendation and intent recognition model training method, device, equipment, medium and product. Background Art

[0002] Content recommendation refers to the process of analyzing various data to screen out some content that meets user interests, needs and preferences from massive content resources (such as articles, videos, products, music, etc.), and displaying these contents to users. Summary of the invention

[0003] The present invention provides a content recommendation and intent recognition model training method, device, equipment, medium and product.

[0004] According to one aspect of the present disclosure, a content recommendation method is provided, comprising: determining a current intention of a target object; in response to determining that the current intention is an extended intention, obtaining associated content of an entry content; the extended intention is used to indicate that the target object is more satisfied with the associated content than with non-associated content; and recommending the associated content to the target object.

[0005] According to another aspect of the present disclosure, a method for training an intent recognition model is provided, comprising: using an intent recognition model to process sample content of a target object to obtain a predicted intent of the target object; constructing a loss function based on the predicted intent and the true intent of the target object; and adjusting model parameters of the intent recognition model based on the loss function to obtain a trained intent recognition model.

[0006] According to another aspect of the present disclosure, a content recommendation device is provided, including: a determination module, used to determine the current intention of a target object; an acquisition module, used to acquire associated content of an entry content in response to determining that the current intention is an extended intention; the extended intention is used to indicate that the target object is more satisfied with the associated content than with non-associated content; and a recommendation module, used to recommend the associated content to the target object.

[0007] According to another aspect of the present disclosure, there is provided an intent recognition model training device, comprising: a prediction module, used to process sample content of a target object using an intent recognition model to obtain a predicted intent of the target object; a construction module, used to construct a loss function based on the predicted intent and the true intent of the target object; and an adjustment module, used to adjust model parameters of the intent recognition model based on the loss function to obtain a trained intent recognition model.

[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute any method as described in any of the above aspects.

[0009] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any one of the methods according to any one of the above aspects.

[0010] According to another aspect of the present disclosure, a computer program product is provided, including a computer program, wherein when the computer program is executed by a processor, the computer program implements any one of the methods described in any one of the above aspects.

[0011] According to the embodiments of the present disclosure, the recommendation accuracy can be improved.

[0012] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a schematic diagram according to a first embodiment of the present disclosure;

[0015] Figure 2 is a schematic diagram of an implementation system for implementing the content recommendation method of an embodiment of the present disclosure;

[0016] Figure 3 is a schematic diagram of an APP interface provided according to an embodiment of the present disclosure;

[0017] Figure 4 is a schematic diagram of a recommendation system provided according to an embodiment of the present disclosure;

[0018] Figure 5 is a schematic diagram of a content recommendation model provided according to an embodiment of the present disclosure;

[0019] Figure 6 is a schematic diagram of an intent recognition model provided according to an embodiment of the present disclosure;

[0020] Figure 7 is a schematic diagram according to a second embodiment of the present disclosure;

[0021] Figure 8is a schematic diagram according to a third embodiment of the present disclosure;

[0022] Fig. 9 is a schematic diagram according to a fourth embodiment of the present disclosure;

[0023] Fig.10 is a schematic diagram according to a fifth embodiment of the present disclosure;

[0024] Fig.11 It is a schematic diagram of an electronic device used to implement the content recommendation method or intent recognition model training method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0026] In the related art, content recommendations are usually made based on the target object's (eg user's) own information, such as the types of content browsed in the past. For example, if the user frequently browses game content, it is easier to push game content.

[0027] Different from the search scenario, the current recommendation scenario usually does not consider the entry content information. The entry content refers to the content clicked by the user on the content list page.

[0028] For this reason, in current recommendation scenarios, the recommended content is usually not related to the entry content.

[0029] Due to different user needs, some users prefer to watch content related to the entry content. The current recommendation solution is difficult to meet the needs of these users and lacks accuracy.

[0030] In order to improve the accuracy of recommendation, the present disclosure provides the following embodiments.

[0031] Figure 1 is a schematic diagram according to the first embodiment of the present disclosure. This embodiment provides a content recommendation method, the method comprising:

[0032] 101. Determine the target audience’s current intentions.

[0033] 102. In response to determining that the current intent is an extended intent, obtain associated content of the entry content; the extended intent is used to indicate that the target object is more satisfied with the associated content than with the non-associated content.

[0034] 103. Recommend the related content to the target object.

[0035] The target object is the recipient of the content recommendation, such as a user.

[0036] In the recommendation scenario, multiple contents can be displayed on the list page. Users can click on one of the contents to jump to the recommendation page according to their needs. On the recommendation page, users can refresh the displayed content through refresh commands (such as swiping up commands). The recommended content displayed on the recommendation page is usually video content. With the development of technology, other styles (or genres) of content have emerged, such as dynamic content, graphic content, etc. For this reason, multiple styles of content can be displayed on the recommendation page, and the corresponding recommendation page can be called a fusion page.

[0037] Entry content refers to the content that users click on on the list page, such as a clip from a film or TV series.

[0038] That is, after the user clicks the entry content on the list page, he / she jumps to the fusion page, and the recommended content is displayed on the fusion page. The user can refresh the recommended content in the fusion page.

[0039] In a search scenario, search results are related to the search term (query). However, in the current recommendation scenario, although the recommended content is displayed based on the triggering of the entry content, the recommended content is usually not related to the entry content.

[0040] In current recommendation scenarios, content recommendations are usually made based on user historical behaviors. Since different users usually have different historical behaviors, different content can be recommended to different users. The recommended content can be called personalized content.

[0041] Content related to the entry content may be referred to as associated content.

[0042] In order to meet the needs of different users, the user's current intention can be identified and personalized content or related content can be recommended based on the current intention.

[0043] The current intent includes: extended intent and non-extended intent, and the extended intent is used to indicate that the target object is more satisfied with the associated content of the entry content than with the non-associated content.

[0044] Taking the example that the non-related content is the usual personalized content, the extended intent indicates that the user's satisfaction with the related content of the entry content is higher than the satisfaction with the personalized content. The non-extended intent can also be called personalized intent, indicating that the user's satisfaction with the personalized content is higher than the satisfaction with the related content of the entry content.

[0045] After determining that the current intention of the target object is an extended intention, the associated content of the entry content is obtained, and the associated content is recommended to the target object.

[0046] On the contrary, if the current intent is non-extended intent, obtain and recommend personalized content.

[0047] Taking the target object as a user as an example, assuming that the current intention of user A is an extended intention, it means that user A is more satisfied with the associated content of the entry content than with the personalized content. In this case, the associated content is recommended to user A. On the contrary, if the current intention of user A is a non-extended intention, it means that user A is more satisfied with the personalized content than with the associated content. In this case, the personalized content is recommended to user A.

[0048] In this embodiment, by identifying the current intent of the target object and performing content recommendations based on the current intent, different types of content can be recommended based on different intentions, thereby improving the accuracy of recommendations. When the current intent is an extended intent, related content of the entry content is recommended to the target object, and related content of the entry content can be recommended, thereby enriching the types of recommended content and improving user experience.

[0049] In order to better understand the present disclosure, the application scenarios involved in the present disclosure are described as follows:

[0050] Figure 2 It is a schematic diagram of an implementation system for implementing the content recommendation method of an embodiment of the present disclosure.

[0051] like Figure 2 As shown, the implementation system includes: a user terminal 201 and a server 202.

[0052] The user terminal 201 may include: a personal computer (PC), a mobile device (such as a mobile phone), a tablet computer, a laptop computer, a smart wearable device, etc.

[0053] The server 202 may be a local server, a cloud server, a single server or a cluster server. The user terminal and the server communicate via a communication network.

[0054] An application (APP) is deployed on the user terminal 201 , and the APP has a content recommendation function. A server corresponding to the APP is deployed on the server 202 .

[0055] After the user opens the APP, the server can execute the content recommendation process, obtain the recommended content, push the recommended content to the APP, and display it to the user through the APP for the user to watch.

[0056] Based on different users, the recommended content can be specific personalized content or related content of the entry content to meet the different needs of users and improve the user experience.

[0057] Figure 3 It is a schematic diagram of an APP interface provided according to an embodiment of the present disclosure.

[0058] like Figure 3 As shown, the APP can be a search engine APP. The page that the user enters after opening the APP can be called a list page. The list page can display a search box and multiple contents for the user to choose from. The user can enter the recommendation page by clicking on a certain content.

[0059] The content that the user clicks on the list page can be called the entry content. When the user clicks on the entry content, he / she will jump to the recommendation page. The recommendation page can integrate multiple styles of content, so it can also be called a fusion page. The fusion page initially displays the entry content, and the user can refresh the recommended content displayed in the fusion page by swiping up or other instructions.

[0060] Due to different user preferences, some users may want to watch personalized content, while some users may want to watch related content. Therefore, it is necessary to improve the accuracy of recommendations to meet the needs of different users.

[0061] Figure 4 is a schematic diagram of a recommendation system provided according to an embodiment of the present disclosure.

[0062] like Figure 4 As shown, the recommendation system as a whole includes: an intention recognition model 401, a first content recommendation model 402 and a second content recommendation model 403.

[0063] The intention recognition model 401 is used to determine the current intention of the target object, where the current intention includes: extended intention or non-extended intention; and when it is determined that the current intention is an extended intention, the first content recommendation model is triggered to perform a content recommendation operation, or when it is determined that the current intention is a non-extended intention, the second content recommendation model is triggered to perform a content recommendation operation.

[0064] The first content recommendation model 402 is used to obtain the related content of the entry content and recommend it to the target object.

[0065] The second content recommendation model 403 is used to obtain personalized content and recommend it to the target object.

[0066] Figure 5 It is a schematic diagram of a content recommendation model provided according to an embodiment of the present disclosure.

[0067] Regardless of whether it is the first content recommendation model or the second content recommendation model, Figure 5 As shown, the whole includes: a recall layer 501, a sorting layer 502, a fusion layer 503 and a distribution layer 504.

[0068] The recall layer 501 mainly selects the content that the user may be interested in from the massive content, such as obtaining N1 (positive integer) pieces of content from the massive content. The N1 pieces of content usually include multiple styles of content, such as video content, dynamic content, and graphic content.

[0069] The sorting layer 502 mainly sorts and filters the N1 pieces of content obtained by the recall layer to obtain N2 (a positive integer, less than N1) pieces of content. These N2 pieces of content usually include multiple styles of content, such as video content, dynamic content, and graphic content.

[0070] The convergence layer 503 mainly determines the preference ratio of each style of content, such as the preference ratio of video content, dynamic content and graphic content is (0.6, 0.3, 0.1), and selects N3 (a positive integer, less than N2) pieces of content from the above-mentioned N2 pieces of content based on the preference ratio, such as obtaining N3*0.6 pieces of video content, N3*0.3 pieces of dynamic content and 0.1*N3 pieces of graphic content.

[0071] The reordering layer 504 mainly selects N4 (a positive integer, less than N3) pieces of content from the N3 pieces of content obtained by the convergence layer for push. Generally speaking, multiple pieces of content, such as 6 pieces of content, are pushed each time.

[0072] The distribution layer 505 is used to determine one of the multiple (N4) pieces of content obtained in the reordering layer and push it to the user when the user generates a trigger instruction (such as a swipe up instruction). Generally speaking, the reordering layer will determine the order of the N4 pieces of content, and the distribution layer can push based on the order.

[0073] The main difference between the first content recommendation model and the second content recommendation model lies in the difference in candidate content. Among them, the N1 pieces of content obtained by the second content recommendation model at the recall layer are mainly personalized content obtained based on the user's historical behavior. The subsequent sorting layer processes these personalized contents, and then recommends content including personalized content.

[0074] The N1 pieces of content obtained by the first content recommendation model at the recall layer include the associated content of the entry content, and the subsequent sorting layer processes these associated contents, and then recommends content including the associated content of the entry content. Specifically, the associated content can be obtained based on preset related rules such as semantic relevance, collaborative relevance, event relevance, and IP relevance.

[0075] In this way, if the current intention of the target object is an extended intention, the associated content of the entry content is recommended to the target object, otherwise the personalized content is recommended in the usual way.

[0076] The current intent of the target object can be determined based on the intent recognition model.

[0077] Figure 6 It is a schematic diagram of an intent recognition model provided according to an embodiment of the present disclosure.

[0078] like Figure 6 As shown, the intent recognition model 601 includes a multilayer perceptron (MLP) and a normalization layer. The MLP is a feedforward neural network model, and the specific structure can be set according to actual needs; the normalization layer can use a sigmoid function to normalize the output of the MLP.

[0079] The intent recognition model processes the input features and outputs the intent score. Then, when the intent score is greater than a preset value, it is determined that the current intent is an extended intent, otherwise it is a non-extended intent.

[0080] The input features may specifically include: object features of the target object, list page click features, fusion page satisfaction features, entry features, etc.

[0081] Taking the target object as a user as an example, the object feature can be obtained by vectorizing user-related data (such as age, gender, historical behavior, etc.).

[0082] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0083] The list page click feature is obtained based on the user's historical click content on the list page. For example, the historical click content is obtained based on the user's historical click behavior on the list page, and the content features of these contents are extracted. For example, the content features of the historical click content are represented by x1~x20 respectively. After encoding these content features, the encoding features of each content are obtained, represented by h1~h20; then, these encoding features (h1~h20) can be fused, such as weighted summation or embedding processing, to obtain the list page click feature.

[0084] The satisfaction feature of the fused page is obtained based on the user's historical satisfaction content on the fused page. For example, the historical satisfaction content can be obtained based on a preset satisfaction index (such as a stay time greater than a preset time). After obtaining the historical satisfaction content, the content features of these contents are extracted. For example, the content features of the historical satisfaction content are represented by s1 to s20 respectively. After encoding these content features, the encoding features of each content are obtained, represented by c1 to c20. Afterwards, these encoding features (c1 to c20) can be fused, such as by weighted summation or embedding processing, to obtain the satisfaction feature of the fused page.

[0085] The entry feature is obtained by vectorizing the relevant information of the entry content. For example, the entry content is embedded to obtain content embedding information; the category of the entry content is vectorized to obtain category features; the genre of the entry content (such as video, dynamic or graphic) is vectorized to obtain genre features; basic features such as duration features can also be obtained. After obtaining these features, these features can be fused, such as weighted summation or embedding processing, to obtain the entry feature.

[0086] The above-mentioned object features, list page click features, fusion page satisfaction features and entry features are obtained, these features are spliced ​​and input into the MLP. The output is the intention score, and the current intention is determined based on the intention score.

[0087] Afterwards, relevant content or personalized content can be recommended to the target object based on the current intent.

[0088] In combination with the above application scenarios, the present disclosure also provides the following embodiments.

[0089] Figure 7 is a schematic diagram according to a second embodiment of the present disclosure. This embodiment provides a content recommendation method, which includes:

[0090] 701. In response to receiving a click instruction from the target object on the entry content, determine the current intention of the target object.

[0091] Take the target object as an example, refer to Figure 3 The page that users enter when opening the APP is called the list page, and the content that users click on the list page is called the entry content.

[0092] After the target object clicks the entry content, the current intention of the target object can be identified.

[0093] In this embodiment, since clicking on the entry content will lead to the recommendation page, the current intention of the target object is determined based on the click instruction of the entry content, so that the current intention can be determined in time, thereby improving the matching of the recommended content displayed on the recommendation page with the current intention and improving the accuracy of the recommendation.

[0094] Specifically, a preset intention recognition model may be used to process input features to output an intention score; the input features include: object features of the target object;

[0095] Based on the intent score, the current intent is determined.

[0096] For example, see Figure 6, the input features are input into the intent recognition model, and the output is the intent score. After that, when the intent score is greater than the preset value, it is determined that the current intent is an extended intent, otherwise it is a non-extended intent.

[0097] In this embodiment, the current intent is determined based on the intent recognition model. Since the intent recognition model is a deep learning model, the excellent performance of the deep learning model can be utilized to efficiently and accurately determine the current intent. Since the input features include object features, different intentions can be identified based on different target objects, thereby improving the personalization of the recommended content.

[0098] In some embodiments, the input feature further includes at least one of the following items:

[0099] The entry characteristics of the entry content;

[0100] List page click features, the list page click features are obtained based on the historical click content of the target object in the list page where the entry content is located;

[0101] A fusion page satisfaction feature, wherein the fusion page satisfaction feature is obtained based on the historical satisfaction content of the target object in the fusion page where the associated content is located.

[0102] In this embodiment, the input features include object features, list page click features, fusion page satisfaction features, entry features, etc. This can enrich the types of input features and improve processing accuracy.

[0103] 702. Determine whether the current intention is an extended intention, if so, execute 703, otherwise execute 704.

[0104] 703. Obtain related content of the entry content, and recommend the related content to the target object.

[0105] Among them, reference Figure 4 , when the current intention is an extended intention, the first content recommendation model is used to recommend related content of the entry content.

[0106] Specifically, the first content recommendation model can be improved from the recall layer. The content recalled by the recall layer includes related content of the entry content. Specifically, the related content can be obtained based on preset related rules such as semantic relatedness, collaborative relatedness, event relatedness, and IP relatedness.

[0107] In this way, a variety of related content can be recalled through a variety of related rules, thereby improving the comprehensiveness and richness of the candidate content, and further improving the comprehensiveness and richness of the recommended content.

[0108] In some embodiments, the associated content may be recommended to the target object based on a predetermined display order; the display order of the associated content may be adjacent to or not adjacent to the display order of the entry content.

[0109] Among them, reference Figure 5 , the reordering layer can push multiple pieces of content to the distribution layer at one time, such as 6 pieces of content, and determine the display order of these contents.

[0110] The display order may also be referred to as display slots. For example, if the first slot is content A and the second slot is content B in the order from first to last, then content A may be displayed first and then content B.

[0111] Specifically, the distribution layer can display these contents one by one based on the display order according to the user's refresh instruction, such as displaying the entrance content initially, displaying the content A of the first slot after the user refreshes once, and displaying the content B of the second slot after refreshing again.

[0112] The multiple pieces of content pushed at one time mentioned above include related content of the entry content, such as content A is related content.

[0113] In the related art, related content is usually not displayed continuously when content is recommended, that is, related content usually needs to be limited to be non-adjacent.

[0114] In this embodiment, the display order of related content is not limited, and the display order of the entry content and its associated content can be adjacent or non-adjacent, that is, the associated content and the entry content can be displayed continuously or discontinuously, such as initially displaying the entry content, and then displaying the associated content with the entry content after refreshing once.

[0115] In this embodiment, by making the display order of the associated content adjacent or non-adjacent to the display order of the entry content, the usual display order limitation of the associated content can be removed, thereby ensuring the timely display of the associated content and improving the user experience.

[0116] 704. Acquire non-related content, and recommend the non-related content to the target object.

[0117] Among them, the non-associated content can be personalized content in the usual plan. When the current intention is not an extended intention, the personalized content can be recommended to the user according to the usual plan.

[0118] In this embodiment, when the current intent is a non-extended intent, non-related content is recommended, which can meet the needs of different users and improve the user experience.

[0119] Figure 8 is a schematic diagram according to the third embodiment of the present disclosure. This embodiment provides an intent recognition model training method, the method comprising:

[0120] 801. Use an intent recognition model to process sample content of a target object to obtain a predicted intent of the target object.

[0121] 802. Construct a loss function based on the predicted intent and the actual intent of the target object.

[0122] 803. Based on the loss function, adjust the model parameters of the intent recognition model to obtain a trained intent recognition model.

[0123] The sample content may include the content of the list page and / or the content of the recommendation page, and the style may include at least one style content, such as one or more of video content, dynamic content, and graphic content.

[0124] Taking the target object as the user, we can collect the user's historical click content on the list page and the historical satisfaction content on the recommendation page. After feature extraction and fusion of these contents, we can get the list page click features and fusion page satisfaction features. In addition, we can also get entry features and object features. After splicing these features, we input them into the intent recognition model, and the output is the predicted intent.

[0125] Predicted intent and true intent can be represented by scores, which are called predicted score and true score respectively. The true score can be marked in advance. If the user is more satisfied with the associated content of the entry content, the true score is represented by 1. If the user is more satisfied with the personalized content, the true score is represented by 0. The predicted score is obtained by the intent recognition model, usually a value between 0 and 1, such as 0.8.

[0126] Taking the predicted score and the true score as an example, a loss function can be constructed based on these two scores, and the loss function can be used to adjust the model parameters to minimize the error between the predicted score and the true score.

[0127] The trained intent recognition model is used to determine the current intent of the target object in the above content recommendation process.

[0128] In this embodiment, by training the intent recognition model, the intent recognition model can be used to determine the current intent of the target object when recommending content, and then recommend different types of content based on different current intentions, thereby improving the accuracy of recommendations.

[0129] The above sample content may include positive samples and negative samples.

[0130] Specifically, recommended content can be displayed to the target object, and the recommended content includes: associated content and personalized content of the entry content; based on the target object's satisfaction with the associated content and the personalized content, positive samples and negative samples are obtained; the sample content includes: the positive sample and the negative sample.

[0131] For example, when multiple recommended contents are pushed to a user for the first time, the recommended contents include related contents and personalized contents. If the user's satisfaction with the related contents is higher than that with the personalized contents, then the sample corresponding to the user is a positive sample, otherwise it is a negative sample.

[0132] Satisfaction can be measured by a preset satisfaction index, such as the dwell time. If the user's dwell time on the content is longer than the preset time, it indicates that the user is satisfied with the content. Assuming that there are 6 recommended contents, 3 related contents that the user is satisfied with, and 1 personalized content that the user is satisfied with, since the satisfaction rate of related contents (3 / 6) is higher than the satisfaction rate of personalized contents (1 / 6), the sample content corresponding to the user is a positive sample.

[0133] In this embodiment, based on the target object's satisfaction with the associated content and the personalized content, positive samples and negative samples are obtained, which can accurately distinguish between positive and negative samples and improve the accuracy of the intent recognition model.

[0134] Fig. 9 is a schematic diagram according to a fourth embodiment of the present disclosure. This embodiment provides a content recommendation device. The device 900 includes: a determination module 901 , an acquisition module 902 and a recommendation module 903 .

[0135] The determination module 901 is used to determine the current intention of the target object; the acquisition module 902 is used to obtain the associated content of the entry content in response to determining that the current intention is an extended intention; the extended intention is used to indicate that the target object is more satisfied with the associated content than with the non-associated content; the recommendation module 903 is used to recommend the associated content to the target object.

[0136] In this embodiment, by identifying the current intent of the target object and performing content recommendations based on the current intent, different types of content can be recommended based on different intentions, thereby improving the accuracy of recommendations. When the current intent is an extended intent, related content of the entry content is recommended to the target object, and related content of the entry content can be recommended, thereby enriching the types of recommended content and improving user experience.

[0137] In some embodiments, the determining module 901 is further configured to:

[0138] Using a preset intention recognition model, the input features are processed to output an intention score; the input features include: object features of the target object;

[0139] Based on the intent score, the current intent is determined.

[0140] In this embodiment, the current intent is determined based on the intent recognition model. Since the intent recognition model is a deep learning model, the excellent performance of the deep learning model can be utilized to efficiently and accurately determine the current intent. Since the input features include object features, different intentions can be identified based on different target objects, thereby improving the personalization of recommended content.

[0141] In some embodiments, the input feature further includes at least one of the following items:

[0142] The entry characteristics of the entry content;

[0143] List page click features, the list page click features are obtained based on the historical click content of the target object in the list page where the entry content is located;

[0144] A fusion page satisfaction feature, wherein the fusion page satisfaction feature is obtained based on the historical satisfaction content of the target object in the fusion page where the associated content is located.

[0145] In this embodiment, the input features include object features, list page click features, fusion page satisfaction features, entry features, etc. This can enrich the types of input features and improve processing accuracy.

[0146] In some embodiments, the determining module 901 is further configured to:

[0147] In response to receiving a click instruction from the target object on the entry content, a current intention of the target object is determined.

[0148] In this embodiment, since clicking on the entry content will lead to the recommendation page, the current intention of the target object is determined based on the click instruction of the entry content, so that the current intention can be determined in time, thereby improving the matching of the recommended content displayed on the recommendation page with the current intention and improving the accuracy of the recommendation.

[0149] In some embodiments, the recommendation module 903 is further used to:

[0150] Based on a predetermined display order, the associated content is recommended to the target object; the display order of the associated content is adjacent to or not adjacent to the display order of the entry content.

[0151] In this embodiment, by making the display order of the associated content adjacent or non-adjacent to the display order of the entry content, the usual display order limitation of the associated content can be removed, thereby ensuring the timely display of the associated content and improving the user experience.

[0152] In some embodiments, the apparatus 900 further includes:

[0153] A presentation module is used for acquiring the non-associated content in response to determining that the current intention is a non-extended intention; and recommending the non-associated content to the target object.

[0154] In this embodiment, when the current intent is a non-extended intent, non-related content is recommended, which can meet the needs of different users and improve the user experience.

[0155] Fig.10 It is a schematic diagram according to the fifth embodiment of the present disclosure. This embodiment provides an intention recognition model training device, and the device 1000 includes: a prediction module 1001, a construction module 1002 and an adjustment module 1003.

[0156] The prediction module 1001 is used to process the sample content of the target object using the intent recognition model to obtain the predicted intent of the target object; the construction module 1002 is used to construct a loss function based on the predicted intent and the true intent of the target object; the adjustment module 1003 is used to adjust the model parameters of the intent recognition model based on the loss function to obtain the trained intent recognition model.

[0157] The trained intent recognition model is used to determine the current intent of the target object in the above content recommendation process.

[0158] In this embodiment, by training the intent recognition model, the intent recognition model can be used to determine the current intent of the target object when recommending content, and then recommend different types of content based on different current intentions, thereby improving the accuracy of recommendations.

[0159] In some embodiments, the apparatus 1000 further includes:

[0160] An acquisition module is used to display recommended content to the target object, and the recommended content includes: associated content and personalized content of the entry content; and based on the target object's satisfaction with the associated content and the personalized content, positive samples and negative samples are acquired; the sample content includes: the positive sample and the negative sample.

[0161] In this embodiment, based on the target object's satisfaction with the associated content and the personalized content, positive samples and negative samples are obtained, which can accurately distinguish between positive and negative samples and improve the accuracy of the intent recognition model.

[0162] It can be understood that in the embodiments of the present disclosure, the same or similar contents in different embodiments can be referenced to each other.

[0163] It can be understood that the “first”, “second”, etc. in the embodiments of the present disclosure are only used for distinction and do not indicate the degree of importance, time sequence, etc.

[0164] It is understandable that unless there is any special limitation on the sequence of steps in the process, it means that the timing relationship between these steps is not limited.

[0165] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0166] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0167] Fig.11 A schematic block diagram of an example electronic device 1100 that can be used to implement an embodiment of the present disclosure is shown. The electronic device 1100 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0168] like Fig.11 As shown, the electronic device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a random access memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the electronic device 1100 can also be stored. The computing unit 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0169] Multiple components in the electronic device 1100 are connected to the I / O interface 1105, including: an input unit 1106, such as a keyboard, a mouse, etc.; an output unit 1107, such as various types of displays, speakers, etc.; a storage unit 1108, such as a disk, an optical disk, etc.; and a communication unit 1109, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 1109 allows the electronic device 1100 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0170] The computing unit 1101 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1101 performs the various methods and processes described above, such as a content recommendation method or an intent recognition model training method. For example, in some embodiments, the content recommendation method or the intent recognition model training method may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1108. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1100 via the ROM 1102 and / or the communication unit 1109. When the computer program is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the content recommendation method or the intent recognition model training method described above may be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured to execute the content recommendation method or the intent recognition model training method in any other appropriate manner (for example, by means of firmware).

[0171] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0172] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable task processing device, so that the program code, when executed by the processor or controller, implements the functions / operations specified in the flow chart and / or block diagram. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0173] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0174] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0175] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0176] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0177] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0178] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A content recommendation method, comprising: Determine the target audience’s current intentions; In response to determining that the current intent is an extended intent, obtaining content associated with the entry content; The extended intention is used to indicate that the target object's satisfaction with the associated content is higher than that with the non-associated content; The associated content is recommended to the target object.

2. The method according to claim 1, wherein: The determining the current intention of the target object includes: Using a preset intention recognition model, the input features are processed to output an intention score; the input features include: object features of the target object; Based on the intent score, the current intent is determined.

3. The method according to claim 2, wherein: The input feature also includes at least one of the following items: The entry characteristics of the entry content; List page click features, the list page click features are obtained based on the historical click content of the target object in the list page where the entry content is located; A fusion page satisfaction feature, wherein the fusion page satisfaction feature is obtained based on the historical satisfaction content of the target object in the fusion page where the associated content is located.

4. The method according to claim 1, wherein: The determining the current intention of the target object includes: In response to receiving a click instruction from the target object on the entry content, a current intention of the target object is determined.

5. The method according to claim 1, wherein: The recommending the related content to the target object includes: Based on a predetermined display order, the associated content is recommended to the target object; the display order of the associated content is adjacent to or not adjacent to the display order of the entry content.

6. The method according to claim 1, further comprising: In response to determining that the current intent is a non-extended intent, acquiring the non-associated content; The non-related content is recommended to the target object.

7. A method for training an intent recognition model, comprising: Using an intention recognition model, processing the sample content of the target object to obtain the predicted intention of the target object; Constructing a loss function based on the predicted intent and the real intent of the target object; Based on the loss function, the model parameters of the intent recognition model are adjusted to obtain a trained intent recognition model.

8. The method according to claim 7, further comprising: Displaying recommended content to the target object, wherein the recommended content includes: related content of the entry content and personalized content; Based on the target object's satisfaction with the associated content and the personalized content, a positive sample and a negative sample are obtained; the sample content includes: the positive sample and the negative sample.

9. The method according to claim 7 or 8, wherein: The intent recognition model is used to determine the current intent in the method according to any one of claims 1-6.

10. A content recommendation device, comprising: A determination module, used to determine the current intention of the target object; an acquisition module, configured to acquire associated content of the entry content in response to determining that the current intent is an extended intent; The extended intention is used to indicate that the target object's satisfaction with the associated content is higher than that with the non-associated content; A recommendation module is used to recommend the related content to the target object.

11. The device according to claim 10, wherein: The determination module is further used for: Using a preset intention recognition model, the input features are processed to output an intention score; the input features include: object features of the target object; Based on the intent score, the current intent is determined.

12. The device according to claim 11, wherein: The input feature also includes at least one of the following items: The entry characteristics of the entry content; List page click features, the list page click features are obtained based on the historical click content of the target object in the list page where the entry content is located; A fusion page satisfaction feature, wherein the fusion page satisfaction feature is obtained based on the historical satisfaction content of the target object in the fusion page where the associated content is located.

13. The device according to claim 10, wherein: The determination module is further used for: In response to receiving a click instruction from the target object on the entry content, a current intention of the target object is determined.

14. The device according to claim 10, wherein: The recommendation module is further used to: Based on a predetermined display order, the associated content is recommended to the target object; the display order of the associated content is adjacent to or not adjacent to the display order of the entry content.

15. The apparatus according to claim 10, further comprising: A presentation module, configured to obtain the non-associated content in response to determining that the current intent is a non-extended intent; And, recommending the non-related content to the target object.

16. An intention recognition model training device, comprising: A prediction module, used to process the sample content of the target object using an intention recognition model to obtain the predicted intention of the target object; A construction module, used to construct a loss function based on the predicted intent and the real intent of the target object; An adjustment module is used to adjust the model parameters of the intent recognition model based on the loss function to obtain a trained intent recognition model.

17. The apparatus according to claim 16, further comprising: An acquisition module is used to display recommended content to the target object, and the recommended content includes: associated content and personalized content of the entry content; and based on the target object's satisfaction with the associated content and the personalized content, positive samples and negative samples are acquired; the sample content includes: the positive sample and the negative sample.

18. The device according to claim 16 or 17, wherein: The intent recognition model is used to determine the current intent in the method according to any one of claims 1-6.

19. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-9.

21. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9.

Images