Large model-based recommendation method and device, electronic equipment and storage medium

Abstract

The present disclosure provides a large model-based recommendation method and device, electronic equipment and storage medium, relating to the technical field of computers, and in particular to the technical field of artificial intelligence such as deep learning, intelligent recommendation, and large models. The specific implementation scheme is as follows: determining content description information of interest corresponding to the target user; inputting the content to be recommended, the content description information of interest, and the current popular search sentence into a large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation sentence associated with the content to be recommended; obtaining the current behavior characteristics of the target user; and in the case that the current behavior characteristics meet the display conditions of the recommendation card, displaying the recommendation card corresponding to at least one content to be recommended.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to the fields of artificial intelligence technology such as deep learning, intelligent recommendation, and large models. Specifically, it relates to a recommendation method, apparatus, electronic device, and storage medium based on a large model. Background Technology

[0002] With the rapid development of the Internet, various information flow products have emerged. Information flow products distribute content products in a waterfall-style manner. Common information flow products include shopping applications, which distribute goods as their content products; and news applications, which distribute news information as their content products.

[0003] However, current recommendation systems mostly use fixed wording for content recommendation, which is monotonous and leads to poor recommendation performance and low accuracy. Summary of the Invention

[0004] This disclosure provides a recommendation method, apparatus, electronic device, and storage medium based on a large model.

[0005] According to a first aspect of this disclosure, a recommendation method based on a large model is provided, comprising:

[0006] Determine the content description information that the target users are interested in;

[0007] The model inputs the content to be recommended, the description of the content of interest, and the current popular search terms to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended.

[0008] Obtain the current behavioral characteristics of the target user;

[0009] If the current behavioral characteristics meet the display conditions for the recommendation card, at least one recommendation card corresponding to the content to be recommended is displayed.

[0010] According to a second aspect of this disclosure, a recommendation apparatus based on a large model is provided, comprising:

[0011] The determination module is used to determine the description information of the content that the target user is interested in.

[0012] The generation module is used to input the content to be recommended, the description information of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended;

[0013] The acquisition module is used to acquire the current behavioral characteristics of the target user;

[0014] The display module is used to display at least one recommendation card corresponding to the content to be recommended when the current behavioral characteristics meet the display conditions for the recommendation card.

[0015] According to a third aspect of this disclosure, an electronic device is provided, comprising:

[0016] At least one processor; and

[0017] A memory communicatively connected to the at least one processor; wherein,

[0018] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the large model-based recommendation method as described in the first aspect.

[0019] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing the computer to perform the large-model-based recommendation method as described in the first aspect.

[0020] According to a fifth aspect of this disclosure, a computer program product is provided, including computer instructions that, when executed by a processor, implement the steps of the large-model-based recommendation method as described in the first aspect.

[0021] The recommendation method, apparatus, electronic device, and storage medium based on large models disclosed herein have the following beneficial effects:

[0022] In this embodiment, the description information of the content of interest corresponding to the target user is first determined. Then, the content to be recommended, the description information of the content of interest, and the current popular search terms are input into a large model to generate at least one recommendation card corresponding to the content to be recommended. This allows the target user's current behavioral characteristics to be obtained. Finally, if the current behavioral characteristics meet the display conditions for the recommendation card, at least one recommendation card corresponding to the content to be recommended is displayed. Therefore, not only can different recommendation cards be generated for different users for the same content to be recommended, combining the description information of the content of interest and popular search terms of different users, making the generated recommendation cards more aligned with user preferences, but also recommendation cards can be displayed to users based on their current behavior, improving the diversity and accuracy of recommendation formats.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0024] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0025] Figure 1 This is a flowchart illustrating a recommendation method based on a large model according to an embodiment of the present disclosure;

[0026] Figure 2 This is a schematic flowchart illustrating a method for determining description information of content of interest, as provided in an embodiment of the present disclosure.

[0027] Figure 3 This is a schematic diagram illustrating the display of a recommendation card according to an embodiment of the present disclosure;

[0028] Figure 4 This is a schematic diagram illustrating the display of a recommendation card according to an embodiment of the present disclosure;

[0029] Figure 5 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of the present disclosure;

[0030] Figure 6 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of the present disclosure;

[0031] Figure 7 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of the present disclosure;

[0032] Figure 8 This is a schematic diagram of the structure of a click-through rate prediction model provided in an embodiment of the present disclosure;

[0033] Figure 9 This is a schematic diagram of a display interface provided in an embodiment of the present disclosure;

[0034] Figure 10 A schematic diagram illustrating the display of follow-up content provided in one embodiment of this disclosure;

[0035] Figure 11 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of the present disclosure;

[0036] Figure 12 This is a schematic diagram of a recommendation device based on a large model according to an embodiment of the present disclosure;

[0037] Figure 13 This is a block diagram of an electronic device used to implement the large model-based recommendation method of the embodiments of this disclosure. Detailed Implementation

[0038] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0039] This disclosure relates to artificial intelligence technologies such as deep learning, intelligent recommendation, and large models.

[0040] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies, and application systems for simulating, extending, and expanding human intelligence.

[0041] Deep learning (DL) is the process of learning the inherent patterns and hierarchical representations of sample data. The information gained during this learning process greatly aids in interpreting data such as text, images, and sound. The ultimate goal of deep learning is to enable machines to possess analytical and learning capabilities similar to humans, allowing them to recognize data such as text, images, and sound.

[0042] Intelligent recommendation, as an important means of information filtering and personalized services, is based on integrating users' historical behavior data, interests and preferences, and contextual information from multiple dimensions. It uses algorithmic models to automatically analyze and predict users' potential needs or interests, and then recommends content, products, services or social relationships that users may be interested in.

[0043] The large model, also known as the Foundation Model, is a model that extracts knowledge from hundreds of millions of corpora or images, learns, and then produces a large model with hundreds of millions of parameters.

[0044] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0045] The following description, with reference to the accompanying drawings, outlines a recommendation method, apparatus, electronic device, and storage medium based on a large model, representing embodiments of this disclosure.

[0046] It should be noted that the execution entity of the large model-based recommendation method in this embodiment is a large model-based recommendation device. This device can be implemented by software and / or hardware and can be configured in an electronic device, which may include, but is not limited to, a terminal or a server.

[0047] Figure 1This is a flowchart illustrating a recommendation method based on a large model according to an embodiment of this disclosure.

[0048] like Figure 1 As shown, this large model-based recommendation method includes:

[0049] S101, Determine the description information of the content that the target user is interested in.

[0050] Among them, the target users can be those who are to receive content recommendations.

[0051] The description of content of interest may include the type of content of interest (e.g., sports, music, entertainment, etc.) and the emotional category of the content of interest (e.g., happy, touching, etc.).

[0052] In some embodiments, the description information of the content that the target user is interested in is determined based on the target user's profile and historical consumption content.

[0053] In some embodiments, a profile of the target user can be generated based on the user's age, gender, interests, etc.

[0054] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information such as age, gender, and interests have been done with the consent of the users, comply with relevant laws and regulations, and do not violate public order and good morals.

[0055] Among them, historical consumption content can be the titles, content, categories, cover images, videos, etc., of resources that the target users have interacted with and have a high completion rate.

[0056] In some embodiments, the target user's profile, historical consumption content, and prompts can be input into a large model, which will then capture descriptions of the target user's interests.

[0057] Figure 2 This is a schematic diagram illustrating a process for determining descriptive information of content of interest, provided as an embodiment of this disclosure. For example... Figure 2 As shown, the text and prompts related to historical consumption content are first encoded using a language instruction encoder to obtain text features H. q The information from images, videos, and other modalities related to historical consumption content is encoded using a vision encoder and then projected onto the input embedding feature space of the large model to obtain visual features H. v This achieves feature alignment across different modalities, and finally integrates the text features H. q and visual features H vThe input is fed into the language model, and the vector X of the language model's response is obtained. a This allows us to obtain descriptions of the content that the target user is interested in.

[0058] S102, input the content to be recommended, the description of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended.

[0059] The content to be recommended can be content intended to be recommended to users. The number of pieces of content to be recommended can be one or more. This disclosure does not limit this.

[0060] Among them, the currently popular search terms can be those with high search volume.

[0061] In some embodiments, the recommendation card may also include cover content (such as images, videos, etc.) that matches the recommendation message.

[0062] In some embodiments, a recommendation message can be generated based on the content to be recommended, the description of the content of interest, and the current popular search terms. Then, a recommendation card can be generated based on the recommendation message.

[0063] The large model used to generate the recommendation card corresponding to the content to be recommended and the large model used to determine the descriptive information of the content of interest can be the same large model or different large models. This disclosure does not limit this.

[0064] S103, Obtain the current behavioral characteristics of the target user.

[0065] The target user's current behavioral characteristics can include their browsing behavior of the content displayed on the interface. For example, this could include clicking on displayed content, the duration of viewing the content, or continuously scrolling through the interface. This disclosure does not limit these behaviors.

[0066] S104, if the current behavioral characteristics meet the display conditions for the recommendation card, display at least one recommendation card corresponding to the content to be recommended.

[0067] In some embodiments, display conditions may include a high number of times the target user clicks on the displayed content, a long dwell time, high frequency of swiping and page turning, and no clicks.

[0068] Figure 3 This is a schematic diagram illustrating the display of a recommendation card according to an embodiment of the present disclosure, as shown below. Figure 3As shown, the display interface includes recommendation cards and other content besides the recommendation cards. The recommendation cards contain recommendation text and guiding information (such as "Click to expand related content").

[0069] Figure 4 This is a schematic diagram illustrating the display of a recommendation card according to an embodiment of the present disclosure, as shown below. Figure 4 As shown, the display interface includes recommendation cards and other content besides the recommendation cards. The recommendation cards contain recommendation text, cover content, and guidance information (such as "Click to expand related content").

[0070] In this embodiment, the description information of the target user's content interests is first determined. The content to be recommended, the description information of the content interests, and current popular search terms are input into a large model to generate at least one recommendation card corresponding to the content to be recommended. Then, the current behavioral characteristics of the target user are obtained. Finally, if the current behavioral characteristics meet the display conditions for the recommendation card, at least one recommendation card corresponding to the content to be recommended is displayed. Therefore, not only can different recommendation cards be generated for different users for the same content to be recommended, combining the description information of different users' content interests and popular search terms, making the generated recommendation cards more aligned with user preferences, but also recommendation cards can be displayed to users based on their current behavior, improving the diversity and accuracy of recommendation formats.

[0071] Figure 5 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of this disclosure, as shown below. Figure 5 As shown, this large model-based recommendation method includes:

[0072] S501, Determine the description information of the content that the target user is interested in.

[0073] S502, obtain the first interaction information of each candidate content in the candidate content library within the first preset time period.

[0074] Among them, the candidate content in the candidate content library can be content that has already been displayed on the platform.

[0075] The first preset time period can be a week, a month, a day, etc. This disclosure does not limit it.

[0076] The first interactive information can be other users' browsing, clicking, and feedback behaviors on the candidate content.

[0077] S503: Based on the first interaction information of each candidate content, determine the current content to be recommended from the candidate content library.

[0078] In this embodiment of the disclosure, candidate content with frequent interactions can be identified as content to be recommended based on the first interaction information. This frequently interacted candidate content is then recommended to the target user, thereby increasing the probability of the target user clicking the recommendation card and improving recommendation accuracy.

[0079] In some embodiments, the score of each candidate content can be comprehensively determined based on the first interaction information, and then the candidate content with the higher score can be identified as the content to be recommended. In some embodiments, the more clicks a candidate content receives, the higher its score; and the more users are interested in the candidate content, the higher its score.

[0080] S504: Input the content to be recommended, the description of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended.

[0081] S505, obtain the current behavioral characteristics of the target user.

[0082] S506, if the current behavioral characteristics meet the display conditions for the recommended card, display at least one recommended card corresponding to the content to be recommended.

[0083] The specific implementation of steps S504 to S506 can be found in the detailed descriptions in other embodiments of this disclosure, and will not be repeated here.

[0084] In this embodiment, the first interaction information of each candidate content in the candidate content library within a first preset time period is obtained. Then, based on the first interaction information of each candidate content, the current content to be recommended is determined from the candidate content library. The content to be recommended, descriptions of content of interest, and current popular search terms are then input into a large model to generate at least one recommendation card corresponding to the content to be recommended. Finally, if the target user's current behavioral characteristics meet the display conditions of the recommendation card, the recommendation card corresponding to at least one content to be recommended is displayed. Therefore, the content to be recommended can be determined based on the first interaction information of each candidate content, allowing for the recommendation of more popular candidate content to the target user. Furthermore, popular content can be presented to different users in different recommendation formats, making the generated recommendation cards corresponding to popular content more aligned with user preferences and further improving recommendation accuracy.

[0085] Figure 6 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of this disclosure; as shown below. Figure 6 As shown, this large model-based recommendation method includes:

[0086] S601, Determine the description information of the content that the target user is interested in.

[0087] S602, input the content to be recommended, the description information of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended.

[0088] S603, obtain the current behavioral characteristics of the target user.

[0089] The specific implementation of steps S601 to S603 can be referred to the detailed descriptions in other embodiments of this disclosure, and will not be repeated here.

[0090] S604, if the current behavioral characteristics meet the display conditions of the recommended card, determine the type of the currently met display conditions.

[0091] In some embodiments, the type of display conditions may include a first type and a second type.

[0092] In some embodiments, the first type is characterized by positive and active behavioral features, such as a high number of clicks or a long dwell time. The second type is characterized by negative and passive behavioral features, such as frequent swiping and page turning, but no clicking behavior.

[0093] S605, determine the target recommended content based on the type of display conditions that are met.

[0094] In some embodiments, when the display condition is of the first type, the target recommended content is determined based on the content currently consumed by the target user.

[0095] It should be noted that when the target user's behavioral characteristics meet the first type of display condition, it means that the target user is quite interested in the content currently consumed. Therefore, based on the content currently consumed, target recommended content related to the currently consumed content can be selected from the content to be recommended. This makes the final recommended content more in line with the target user's current interests, thereby increasing the probability of the target user clicking on the displayed recommended card and further improving the accuracy of the recommendation.

[0096] In some embodiments, if the display condition that the target user's behavioral characteristics meet is of the first type, the content to be recommended that belongs to the same type as the currently consumed content can be determined as the target recommended content.

[0097] For example, if the current content consumed is music, then the recommended content in the music category will be determined as the target recommended content.

[0098] In some embodiments, if the display condition that the target user's behavioral characteristics meet is of the first type, the content to be recommended that has a similarity greater than a first threshold with the currently consumed content is determined as the target recommended content.

[0099] In some embodiments, the similarity between the content to be recommended and the currently consumed content can be determined by methods such as pre-similarity and Euclidean distance.

[0100] In this embodiment of the disclosure, content that is highly related to the currently consumed content can be identified as target recommended content based on two dimensions: the type of the currently consumed content and the similarity between the content and the currently consumed content. This not only increases the likelihood that the identified target recommended content will be clicked by the target user, thereby improving the accuracy of the recommendation, but also increases the diversity of information related to the currently consumed content.

[0101] In some embodiments, when the display condition is of the second type, the target recommended content is determined based on the content currently displayed on the display interface.

[0102] It should be noted that when the target user's behavioral characteristics meet the second type of display condition, it means that the target user is not interested in the content currently displayed on the interface. Therefore, based on the currently displayed content, target consumption content that is not related to the currently displayed content can be selected from the content to be recommended, and other content that the target user may be interested in can be recommended, thereby increasing the probability of the target user clicking on the displayed recommendation card and thus improving the accuracy of the recommendation.

[0103] In some embodiments, when the display condition that the target user's behavioral characteristics meet is of the second type, the content to be recommended that has a similarity to the currently displayed content that is less than a second threshold is determined as the target recommended content.

[0104] In some embodiments, the second threshold is less than or equal to the first threshold.

[0105] In some embodiments, if the display condition that the target user's behavioral characteristics meet is of the second type, the content to be recommended that does not belong to the same type as the currently displayed content is determined as the target recommended content.

[0106] For example, if the currently displayed content is music, then non-music content to be recommended will be identified as the target recommended content.

[0107] In this embodiment of the disclosure, content that is not related to the currently consumed content can be identified as target recommended content based on two dimensions: the type of the currently displayed content and the similarity between the content and the currently displayed content. This allows for the provision of target recommended content that is of interest to the target user as much as possible, thereby further improving the accuracy of recommendations.

[0108] S606, Display the recommendation card corresponding to the target recommended content.

[0109] In this embodiment, after inputting the content to be recommended, descriptions of content of interest, and current popular search terms into a large model, at least one recommendation card corresponding to the content to be recommended is generated. Each recommendation card contains a recommendation phrase associated with the content to be recommended. Then, if the target user's current behavioral characteristics meet the display conditions of the recommendation card, the type of the currently met display conditions is determined. Based on the type of met display conditions, the target recommended content is determined, and finally, the recommendation card corresponding to the target recommended content is displayed. Therefore, based on the type of display conditions met by the target user's current behavioral characteristics, the target recommended content that the user may be more interested in can be determined, further improving recommendation accuracy.

[0110] Figure 7 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of this disclosure; as shown below. Figure 7 As shown, this large model-based recommendation method includes:

[0111] S701, Determine the description information of the content that the target user is interested in.

[0112] S702, input the content to be recommended, the description information of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended.

[0113] S703, obtain the current behavioral characteristics of the target user.

[0114] S704, if the current behavioral characteristics meet the display conditions for the recommended card, display at least one recommended card corresponding to the content to be recommended.

[0115] The specific implementation of steps S701 to S704 can be referred to the detailed descriptions in other embodiments of this disclosure, and will not be repeated here.

[0116] S705: For each recommended card, determine the click-through rate (CTR) based on a click-through rate prediction model.

[0117] The click-through rate prediction model can be a reward model, and its structure can be a Decoder-only Transformer structure.

[0118] In some embodiments, the recommendation text in the recommendation card, the target user's behavioral sequence during a second preset time period, and the target user's profile are input into the click-through rate prediction model to obtain the click-through rate. Therefore, by combining the target user's profile and behavioral information, the click-through rate of the recommendation card can be predicted more accurately.

[0119] In some embodiments, Figure 8 This is a schematic diagram of the structure of a click-through rate prediction model provided in an embodiment of the present disclosure, as shown below. Figure 8 As shown, the training method for the click-through rate (CTR) prediction model may include: acquiring training data, which includes recommendation samples, user profile samples, user historical behavior sequence samples, and CTR labels; inputting the recommendation samples, user profile samples, user historical behavior sequence samples, and CTR labels into the CTR prediction model to be trained to obtain the predicted CTR and the predicted behavior sequence corresponding to the historical behavior sequence; determining the target loss based on the recommendation loss between the predicted CTR and the CTR labels, and the contrastive loss between the historical behavior sequence and the predicted behavior sequence; and training the CTR prediction model.

[0120] Specifically, the predicted behavior sequence can be obtained by predicting the (i+1)th behavior based on the first i behaviors in the historical behavior sequence. In some embodiments, a masked-transformer can be used to mask the remaining behaviors except for the first i behaviors, thereby enabling the prediction of the (i+1)th behavior based on the first i behaviors in the historical behavior sequence.

[0121] The second preset time period can be the time period before the current time.

[0122] The behavioral sequence corresponding to the second preset time period can be the target user's clicks, feedback, and browsing behavior on the content within the second preset time period.

[0123] In some embodiments, third sample data is determined, which includes click information corresponding to the recommended card and consumption information of the follow-up content of the recommended card. This third sample data is used to update and train the click-through rate (CTR) prediction model. Therefore, the CTR prediction model can be updated based on the target user's click information on the recommended card and the consumption information of the follow-up content, thereby improving the accuracy of the CTR prediction model in predicting CTR.

[0124] S706, obtain the second interaction information corresponding to the recommended card.

[0125] In some embodiments, the second interaction information corresponding to the recommended card is determined based on the click information of the recommended card and the consumption information of the follow-up content of the recommended card, wherein the follow-up content is the content displayed after the recommended card is clicked. Therefore, by combining the two dimensions of interaction information—the click status of the recommended card and the consumption information of the follow-up content—more comprehensive interaction information can be provided to accurately determine the score of the recommended card, thereby improving the accuracy of the determined score.

[0126] The click information for the recommended card can include clicks from the target user, or clicks from other users. Alternatively, it can be the ratio of the number of clicks to the total number of recommendations.

[0127] The consumption information for the follow-up content of the recommended card may include consumption information of the target user on the follow-up content, or consumption information of other users on the follow-up content. This disclosure does not limit this.

[0128] Figure 9 This is a schematic diagram of a display interface provided in one embodiment of the present disclosure, such as... Figure 9 As shown, in Figure 3 or Figure 4 When a recommended card is clicked, a waiting message (such as "fresh content will be presented to you soon") will be displayed on the screen.

[0129] Figure 10 A schematic diagram illustrating the display of follow-up content provided in one embodiment of this disclosure, such as... Figure 10 As shown, after the recommended card is clicked, the display interface will refresh, showing follow-up content 1, follow-up content 2, etc., related to the recommendation.

[0130] S707 determines the score of the recommended card based on click-through rate and second interaction information.

[0131] In some embodiments, a higher click-through rate results in a higher score; the higher the ratio of clicks to recommendations for a recommended card, the higher the score; and the higher the quality of the follow-up content for a recommended card, the higher the score.

[0132] In some embodiments, where the second interactive information includes consumption information of follow-up content, the relevance between the follow-up resource and the recommendation can be determined first based on the consumption information of the follow-up content. Then, based on the consumption information of the follow-up content and the relevance between the follow-up resource and the recommendation, the quality of the follow-up resource is evaluated. Finally, the score of the recommendation card is comprehensively analyzed by combining the click-through rate, the evaluation of the follow-up resource, and the click information corresponding to the recommendation card.

[0133] In some embodiments, clicked follow-up content can be used as positive samples, while unclicked follow-up content can be used as negative samples. The third model can be fine-tuned through contrastive learning to evaluate the relevance between follow-up resources and recommendations. Furthermore, the relevance between follow-up resources and recommendations can be determined using the third model.

[0134] S708, Determine the first sample data, wherein the first sample data includes the score of the recommendation card, the content to be recommended corresponding to the recommendation card, the description information of the content of interest to the target user, and popular search terms. The first sample data is used to update and train the large model.

[0135] In this embodiment of the disclosure, after determining the score corresponding to the recommendation card, the large model can be updated and trained based on the score of the recommendation card, the content to be recommended corresponding to the recommendation card, the description information of the content of interest to the target user, and popular search terms. This can improve the performance of the large model, enabling the generation of recommendation cards that are more of interest to the target user in subsequent recommendation tasks, thereby improving recommendation accuracy.

[0136] In this embodiment of the disclosure, after displaying at least one recommendation card corresponding to content to be recommended, the click-through rate (CTR) of each recommendation card can be determined based on a click-through rate prediction model. Second interaction information corresponding to the recommendation card is obtained. Based on the CTR and the second interaction information, the score of the recommendation card is determined. Finally, first sample data is determined, wherein the first sample data includes the score of the recommendation card, the content to be recommended corresponding to the recommendation card, the description information of the target user's interested content, and popular search terms. The first sample data is used to update and train the large model. Therefore, the quality of the recommendation card can be comprehensively judged from the predicted CTR and the actual interaction information of the recommendation card. Based on the quality of the recommendation card, the content to be recommended corresponding to the recommendation card, the description information of the target user's interested content, and popular search terms, the large model can be guided to update, thereby improving the performance of the large model. This allows for the generation of recommendation cards that are more interesting to the target user in subsequent recommendation tasks, improving recommendation accuracy.

[0137] Figure 11 This is a flowchart illustrating a recommendation method based on a large model according to yet another embodiment of this disclosure; as shown below. Figure 11 As shown, this large model-based recommendation method includes:

[0138] S1101, Determine the description information of the content that the target user is interested in.

[0139] S1102, Input the content to be recommended, the description information of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended.

[0140] S1103, Obtain the current behavioral characteristics of the target user.

[0141] S1104, if the current behavioral characteristics meet the display conditions for the recommendation card, display at least one recommendation card corresponding to the content to be recommended.

[0142] S1105, For each recommended card, determine the click-through rate (CTR) of the recommended card based on the CTR prediction model.

[0143] S1106, Obtain the second interaction information corresponding to the recommended card.

[0144] S1107, Determine the score of the recommended card based on the click-through rate and the second interaction information.

[0145] The specific implementation of steps S1101 to S1107 can be referred to the detailed descriptions in other embodiments of this disclosure, and will not be repeated here.

[0146] S1108, when any content to be recommended corresponds to at least two recommendation cards, determine the recommendation weights corresponding to the at least two recommendation cards based on the scores of each of the at least two recommendation cards.

[0147] In some embodiments, at least two recommendation cards can be generated for the same content to be recommended, wherein the recommendation text in each recommendation card is different.

[0148] In some embodiments, the higher the score of a recommended card, the higher its recommendation weight.

[0149] S1109, Determine the second sample data, wherein the second sample data includes recommendation cards, recommendation weights corresponding to recommendation cards, any content to be recommended, description information of content of interest to the target user, and popular search terms. The second sample data is used to update and train the large model.

[0150] In this embodiment of the disclosure, after determining the recommendation weight corresponding to the recommendation card, the large model can be updated and trained based on any content to be recommended, the recommendation card corresponding to any recommended content, the recommendation weight corresponding to the recommendation card, the description information of the target user's content of interest, and popular search terms. This allows the large model to learn the recommendation weight corresponding to the recommendation card corresponding to any content to be recommended. As a result, when any recommended content is recommended again, the recommendation card with the higher recommendation weight can be displayed to the user first, thereby improving the accuracy of the recommendation.

[0151] In this embodiment of the disclosure, after displaying at least one recommendation card corresponding to content to be recommended, the click-through rate (CTR) of each recommendation card can be determined based on a click-through rate prediction model. Second interaction information corresponding to the recommendation card is obtained. Based on the CTR and the second interaction information, the score of the recommendation card is determined. When any content to be recommended corresponds to at least two recommendation cards, the recommendation weights corresponding to the at least two recommendation cards are determined based on the scores of each of the at least two recommendation cards. Finally, second sample data is determined, which includes recommendation cards, the recommendation weights corresponding to the recommendation cards, any content to be recommended, descriptive information of the target user's interests, and popular search terms. The second sample data is used to update and train the large model. Therefore, the recommendation weights of recommendation cards can be determined based on their scores, guiding the large model to learn the recommendation weights corresponding to recommendation cards under any content to be recommended. Consequently, when any content to be recommended is recommended again, recommendation cards with higher recommendation weights can be prioritized for display to the user, improving recommendation accuracy.

[0152] Figure 12 This is a schematic diagram of a recommendation device based on a large model according to an embodiment of the present disclosure;

[0153] like Figure 12 As shown, the large-model-based recommendation device 1200 includes:

[0154] The determination module 1201 is used to determine the description information of the content of interest to the target user.

[0155] The generation module 1202 is used to input the content to be recommended, the description information of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended;

[0156] Module 1203 is used to acquire the current behavioral characteristics of the target user;

[0157] The display module 1204 is used to display at least one recommendation card corresponding to the content to be recommended when the current behavioral characteristics meet the display conditions of the recommendation card.

[0158] In some embodiments of this disclosure, a first processing module is further included, configured to:

[0159] Obtain the first interaction information of each candidate content in the candidate content library within a first preset time period;

[0160] Based on the first interaction information of each candidate content, the current content to be recommended is determined from the candidate content library.

[0161] In some embodiments of this disclosure, the display module 1204 is used for:

[0162] If the current behavioral characteristics meet the display conditions of the recommended card, determine the type of the currently met display conditions;

[0163] Based on the type of display conditions met, determine the target recommended content;

[0164] Displays the recommendation card corresponding to the target recommended content.

[0165] In some embodiments of this disclosure, the display module 1204 is used for:

[0166] If the display condition meets the first type, the target recommended content is determined based on the content currently consumed by the target user; or,

[0167] If the display condition is of type two, the target recommended content is determined based on the content currently displayed on the display interface.

[0168] In some embodiments of this disclosure, the display module 1204 is used for:

[0169] Content that belongs to the same category as the currently consumed content will be identified as target recommendation content; or...

[0170] Content that has a similarity to currently consumed content greater than a first threshold is identified as target recommendation content.

[0171] In some embodiments of this disclosure, the display module 1204 is used for:

[0172] Content that has a similarity to the currently displayed content less than a second threshold is identified as target recommended content; or...

[0173] Content that does not belong to the same category as the currently displayed content will be identified as target recommended content.

[0174] In some embodiments of this disclosure, a second processing module is further included, for:

[0175] For each recommended card, the click-through rate (CTR) is determined based on a click-through rate prediction model.

[0176] Obtain the second interaction information corresponding to the recommended card;

[0177] The score of the recommended card is determined based on the click-through rate and the second interaction information;

[0178] The first sample data is determined, which includes the score of the recommendation card, the content to be recommended corresponding to the recommendation card, the description information of the content of interest to the target user, and popular search terms. The first sample data is used to update and train the large model.

[0179] In some embodiments of this disclosure, the second processing module is further configured to:

[0180] When any content to be recommended corresponds to at least two recommendation cards, the recommendation weights corresponding to the at least two recommendation cards are determined based on the scores of each recommendation card in the at least two recommendation cards.

[0181] The second sample data is determined, which includes recommendation cards, recommendation weights corresponding to recommendation cards, any content to be recommended, descriptions of content of interest to the target user, and popular search terms. The second sample data is used to update and train the large model.

[0182] In some embodiments of this disclosure, the second processing module is configured to:

[0183] Based on the click information corresponding to the recommended card and the consumption information of the follow-up content of the recommended card, the second interaction information corresponding to the recommended card is determined, wherein the follow-up content is the content displayed after the recommended card is clicked.

[0184] In some embodiments of this disclosure, the second processing module is configured to:

[0185] Input the recommendation message from the recommendation card, the target user's behavior sequence during the second preset time period, and the target user's profile into the click-through rate prediction model to obtain the click-through rate.

[0186] In some embodiments of this disclosure, the apparatus further includes a third processing module for:

[0187] The third sample data is determined, which includes the click information of the recommended card and the consumption information of the follow-up content of the recommended card. The third sample data is used to update and train the click-through rate prediction model.

[0188] It should be noted that the foregoing explanation of the large model-based recommendation method also applies to the large model-based recommendation device of this embodiment, and will not be repeated here.

[0189] In this embodiment, the description information of the target user's content interests is first determined. The content to be recommended, the description information of the content interests, and current popular search terms are input into a large model to generate at least one recommendation card corresponding to the content to be recommended. Then, the current behavioral characteristics of the target user are obtained. Finally, if the current behavioral characteristics meet the display conditions of the recommendation card, at least one recommendation card corresponding to the content to be recommended is displayed. Therefore, not only can different recommendation cards be generated for different users for the same content to be recommended, combining the description information of different users' content interests and popular search terms, making the generated recommendation cards more aligned with user preferences, but also recommendation cards can be displayed to users based on their current behavior, improving the diversity and accuracy of recommendation formats.

[0190] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0191] Figure 13 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0192] like Figure 13 As shown, device 1300 includes a computing unit 1301, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1302 or a computer program loaded from storage unit 1308 into random access memory (RAM) 1303. The RAM 1303 may also store various programs and data required for the operation of device 1300. The computing unit 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. Input / output (I / O) interface 1305 is also connected to bus 1304.

[0193] Multiple components in device 1300 are connected to I / O interface 1305, including: input unit 1306, such as keyboard, mouse, etc.; output unit 1307, such as various types of monitors, speakers, etc.; storage unit 1308, such as disk, optical disk, etc.; and communication unit 1309, such as network card, modem, wireless transceiver, etc. Communication unit 1309 allows device 1300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0194] The computing unit 1301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1301 performs the various methods and processes described above, such as the large model-based recommendation method. For example, in some embodiments, the large model-based recommendation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1308. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1300 via ROM 1302 and / or communication unit 1309. When the computer program is loaded into RAM 1303 and executed by the computing unit 1301, one or more steps of the large model-based recommendation method described above may be performed. Alternatively, in other embodiments, computing unit 1301 may be configured to perform a large-model-based recommendation method by any other suitable means (e.g., by means of firmware).

[0195] Various embodiments 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-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0196] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0197] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0198] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, 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 sound input, voice input, or tactile input).

[0199] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), the Internet, and blockchain networks.

[0200] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0201] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0202] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified. In the description of this disclosure, the words "if" and "suppose" as used may be interpreted as "when," "when," "in response to determination," or "in the circumstances."

[0203] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A recommendation method based on a large model, comprising: Determine the content description information that the target users are interested in; Obtain the first interaction information of each candidate content in the candidate content library within a first preset time period; Based on the first interaction information of each candidate content, the current content to be recommended is determined from the candidate content library; The content to be recommended, the description of the content of interest, and the current popular search terms are input into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended; Obtain the current behavioral characteristics of the target user; wherein, the current behavioral characteristics include the target user's behavior of browsing the content displayed on the display interface; If the current behavioral characteristic satisfies the first type of display condition for the recommendation card, the type of the currently satisfied display condition is determined, and the target recommendation content is determined based on the content currently consumed by the target user. If the display condition is the second type, the target recommendation content is determined based on the content currently displayed on the display interface. The recommendation card corresponding to the target recommendation content is then displayed. Wherein, the first type indicates that the current behavioral characteristic is positive, and the second type indicates that the current behavioral characteristic is negative. For each recommended card, the click-through rate (CTR) of the recommended card is determined based on a click-through rate prediction model; second interaction information corresponding to the recommended card is obtained; the score of the recommended card is determined based on the CTR and the second interaction information; first sample data is determined, wherein the first sample data includes the score of the recommended card, the content to be recommended corresponding to the recommended card, the description information of the content of interest to the target user, and the popular search terms, and the first sample data is used to update and train the large model.

2. The method as described in claim 1, wherein, Determining the target recommended content based on the content currently consumed by the target user includes: Content of the same type as the currently consumed content will be identified as the target recommended content; or... Content that has a similarity to the currently consumed content greater than a first threshold is identified as the target recommended content.

3. The method as described in claim 1, wherein, The step of determining the target recommended content based on the content currently displayed on the display interface includes: Content to be recommended that has a similarity to the currently displayed content less than a second threshold is identified as the target recommended content; or... Content that is not of the same type as the currently displayed content is identified as the target recommended content.

4. The method as described in any one of claims 1-3, wherein, After determining the score of the recommended card, the process further includes: When any content to be recommended corresponds to at least two recommendation cards, the recommendation weight corresponding to each of the at least two recommendation cards is determined based on the score of each of the at least two recommendation cards; A second sample data is determined, wherein the second sample data includes recommendation cards, recommendation weights corresponding to the recommendation cards, any content to be recommended, description information of the content of interest to the target user, and popular search terms. The second sample data is used to update and train the large model.

5. The method as described in any one of claims 1-3, wherein, The step of obtaining the second interaction information corresponding to the recommended card includes: Based on the click information corresponding to the recommended card and the consumption information of the follow-up content of the recommended card, the second interaction information corresponding to the recommended card is determined, wherein the follow-up content is the content displayed after the recommended card is clicked.

6. The method as described in any one of claims 1-3, wherein, The step of determining the click-through rate (CTR) of each recommended card based on a click-through rate prediction model includes: The click-through rate (CTR) is obtained by inputting the recommendation message in the recommendation card, the behavioral sequence of the target user during the second preset time period, and the profile of the target user into the click-through rate prediction model.

7. The method of claim 6, wherein, The method further includes: A third sample data is determined, wherein the third sample data includes the click information corresponding to the recommended card and the consumption information of the follow-up content of the recommended card, and the third sample data is used to update and train the click-through rate prediction model.

8. A recommendation device based on a large model, comprising: The determination module is used to determine the description information of the content that the target user is interested in. The first processing module is used to obtain the first interaction information of each candidate content in the candidate content library within a first preset time period; Based on the first interaction information of each candidate content, the current content to be recommended is determined from the candidate content library; The generation module is used to input the content to be recommended, the description information of the content of interest, and the current popular search terms into the large model to generate at least one recommendation card corresponding to the content to be recommended, wherein the recommendation card contains a recommendation text associated with the content to be recommended; The acquisition module is used to acquire the current behavioral characteristics of the target user; wherein, the current behavioral characteristics include the target user's behavior of browsing the content displayed on the display interface; The display module is configured to: determine the type of display condition currently met when the current behavioral characteristics satisfy the display condition of the recommendation card as a first type; determine the target recommendation content based on the content currently consumed by the target user; and determine the target recommendation content based on the content currently displayed on the display interface when the display condition is a second type. The module also displays the recommendation card corresponding to the target recommendation content. The first type indicates that the current behavioral characteristics are positive, and the second type indicates that the current behavioral characteristics are negative. The second processing module is used to determine the click-through rate (CTR) of each recommended card based on a click-through rate prediction model; obtain the second interaction information corresponding to the recommended card; determine the score of the recommended card based on the CTR and the second interaction information; and determine first sample data, wherein the first sample data includes the score of the recommended card, the content to be recommended corresponding to the recommended card, the description information of the target user's content of interest, and the popular search terms, and the first sample data is used to update and train the large model.

9. The apparatus of claim 8, wherein, The display module is used for: Content of the same type as the currently consumed content will be identified as the target recommended content; or... Content that has a similarity to the currently consumed content greater than a first threshold is identified as the target recommended content.

10. The apparatus of claim 8, wherein, The display module is used for: Content to be recommended that has a similarity to the currently displayed content less than a second threshold is identified as the target recommended content; or... Content that is not of the same type as the currently displayed content is identified as the target recommended content.

11. The apparatus according to any one of claims 8-10, wherein, The second processing module is further configured to: When any content to be recommended corresponds to at least two recommendation cards, the recommendation weight corresponding to each of the at least two recommendation cards is determined based on the score of each of the at least two recommendation cards; A second sample data is determined, wherein the second sample data includes recommendation cards, recommendation weights corresponding to the recommendation cards, any content to be recommended, description information of the content of interest to the target user, and popular search terms. The second sample data is used to update and train the large model.

12. The apparatus according to any one of claims 8-10, wherein, The second processing module is used for: Based on the click information corresponding to the recommended card and the consumption information of the follow-up content of the recommended card, the second interaction information corresponding to the recommended card is determined, wherein the follow-up content is the content displayed after the recommended card is clicked.

13. The apparatus according to any one of claims 8-10, wherein, The second processing module is used for: The click-through rate (CTR) is obtained by inputting the recommendation message in the recommendation card, the behavioral sequence of the target user during the second preset time period, and the profile of the target user into the click-through rate prediction model.

14. The apparatus of claim 13, wherein, The device further includes a third processing module for: A third sample data is determined, wherein the third sample data includes the click information corresponding to the recommended card and the consumption information of the follow-up content of the recommended card, and the third sample data is used to update and train the click-through rate prediction model.

15. 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 to enable the at least one processor to perform the method of any one of claims 1-7.

16. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

17. A computer program product comprising computer instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1-7.

Images