Recommended methods, apparatus, electronic devices, storage media, and program products

Abstract

This disclosure provides a recommendation method, relating to the field of data processing technology, particularly to the fields of natural language processing, artificial intelligence, intelligent search, and intelligent recommendation. The specific implementation scheme is as follows: based on input information, multiple publishing objects matching the input information are determined from a pool of candidate publishing objects; the initial recommendation parameters of each of the multiple publishing objects are updated to obtain target recommendation parameters for each of the multiple publishing objects, where the initial recommendation parameters characterize the recommendation level of the publishing object; and the content to be recommended for each of the multiple publishing objects is filtered using the target recommendation parameters of each of the multiple publishing objects and the input information to determine the target content for recommendation. This disclosure also provides a recommendation device, electronic device, storage medium, and program product.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to the fields of natural language processing, artificial intelligence, intelligent search, and intelligent recommendation, specifically to a recommendation method, apparatus, electronic device, storage medium, and program product. Background Technology

[0002] With the booming development of the internet, information of all kinds has experienced explosive growth. Against this backdrop, recommendation systems have emerged, essentially playing the role of information filtering and selection, improving the efficiency of users' information acquisition. How to provide users with diverse information while meeting their recommendation needs is a direction worth exploring. Summary of the Invention

[0003] This disclosure provides a recommended method, apparatus, electronic device, storage medium, and program product.

[0004] According to one aspect of this disclosure, a recommendation method is provided, comprising: determining multiple publishing objects that match the input information from multiple candidate publishing objects based on input information; updating the initial recommendation parameters of each of the multiple publishing objects to obtain target recommendation parameters of each of the multiple publishing objects, wherein the initial recommendation parameters characterize the recommendation level of the publishing object; and using the target recommendation parameters of each of the multiple publishing objects and the input information to filter the content to be recommended for each of the multiple publishing objects to determine the target content for recommendation.

[0005] According to another aspect of this disclosure, a recommendation apparatus is provided, comprising: a first determining module, configured to determine, based on input information, a plurality of publishing objects that match the input information from a plurality of candidate publishing objects; an updating module, configured to update the initial recommendation parameters of each of the plurality of publishing objects to obtain target recommendation parameters of each of the plurality of publishing objects, wherein the initial recommendation parameters characterize the recommendation level of the publishing object; and a second determining module, configured to use the target recommendation parameters of each of the plurality of publishing objects and the input information to filter the content to be recommended of each of the plurality of publishing objects to determine the target content for recommendation.

[0006] According to another aspect of this 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, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described above.

[0007] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method described above.

[0008] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described above.

[0009] 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

[0010] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0011] Figure 1 This illustration schematically shows an exemplary system architecture to which the recommended methods and apparatus can be applied according to embodiments of this disclosure;

[0012] Figure 2 A flowchart illustrating a recommended method according to an embodiment of this disclosure is shown schematically;

[0013] Figure 3 A flowchart illustrating the determination of the publishing target and target recommendation parameters according to an embodiment of this disclosure is shown schematically;

[0014] Figure 4 A flowchart illustrating a recommended method according to another embodiment of this disclosure is shown schematically;

[0015] Figure 5 The schematic diagram illustrates the principle of determining target content based on an index tree according to an embodiment of the present disclosure;

[0016] Figure 6 A block diagram of a recommended apparatus according to embodiments of the present disclosure is schematically shown; and

[0017] Figure 7 A block diagram of an electronic device suitable for implementing the recommended method according to an embodiment of the present disclosure is illustrated schematically. Detailed Implementation

[0018] 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.

[0019] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information comply with the provisions of relevant laws and regulations, necessary confidentiality measures have been taken, and there is no violation of public order and good morals.

[0020] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0021] Figure 1 The illustration schematically depicts an exemplary system architecture to which recommended methods and apparatus can be applied according to embodiments of the present disclosure.

[0022] It is important to note that Figure 1 The examples shown are merely illustrative of system architectures applicable to embodiments of this disclosure, intended to help those skilled in the art understand the technical content of this disclosure. They do not imply that embodiments of this disclosure cannot be used in other devices, systems, environments, or scenarios. For instance, in another embodiment, an exemplary system architecture to which the recommended methods and apparatus can be applied may include a terminal device. However, the terminal device can implement the recommended methods and apparatus provided in the embodiments of this disclosure without interacting with a server.

[0023] like Figure 1 As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0024] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (for example only).

[0025] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0026] Server 105 can be a server that provides various services, such as a backend management server that supports the content browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0027] It should be noted that the recommended methods provided in the embodiments of this disclosure can generally be executed by terminal devices 101, 102, or 103. Accordingly, the recommended devices provided in the embodiments of this disclosure can also be disposed in terminal devices 101, 102, or 103.

[0028] Alternatively, the recommended method provided in this disclosure can generally be executed by server 105. Correspondingly, the recommended device provided in this disclosure can generally be located in server 105. The recommended method provided in this disclosure can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the recommended device provided in this disclosure can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105.

[0029] For example, when a user searches for information, terminal devices 101, 102, and 103 can acquire the user's input information and then send it to server 105. Server 105 analyzes the input information and recommends target content to the user based on it. Alternatively, a server or server cluster capable of communicating with terminal devices 101, 102, 103, and / or server 105 can analyze the input information and ultimately recommend target content to the user.

[0030] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0031] It should be noted that the sequence numbers of the operations in the following methods are for descriptive purposes only and should not be considered as indicating the execution order of the operations. Unless explicitly stated otherwise, the method does not need to be executed in the exact order shown.

[0032] When recommending relevant content to users, language model-based recommendation systems can use the language model to understand user input information, such as a query, to generate published items and keywords. Leveraging the powerful capabilities of the language model, it recommends relevant content matching the query. However, the training data for language models comes from funnel sampling within the recommendation system. This results in frequent collection of data from high-traffic published items, while data from low-traffic items is relatively scarce. Consequently, in some cases, the results generated by the language model may be concentrated on content from a few high-traffic items. For example, in ad search recommendation scenarios, the ad content generated by the language model tends to concentrate on a few high-traffic advertisers, thus affecting the diversity of the recommendation system's results.

[0033] This disclosure provides a recommendation method, comprising: determining multiple publishing objects that match the input information from multiple candidate publishing objects based on input information; updating the initial recommendation parameters of each of the multiple publishing objects to obtain target recommendation parameters of each of the multiple publishing objects, wherein the initial recommendation parameters characterize the recommendation level of the publishing object; and using the target recommendation parameters of each of the multiple publishing objects and the input information to filter the content to be recommended for each of the multiple publishing objects to determine the target content for recommendation.

[0034] According to the recommendation method disclosed herein, target recommendation parameters for each of the multiple publishing objects are obtained by updating their respective recommendation levels. Based on the target recommendation parameters and the search content, target content for recommendation is determined. Thus, by adjusting the recommendation levels of the multiple publishing objects, the diversity of subsequent target content for recommendation is improved while ensuring the recommendation success rate.

[0035] Figure 2 A flowchart illustrating a recommended method according to an embodiment of this disclosure is shown schematically.

[0036] like Figure 2 As shown, the method includes operations S210~S230.

[0037] In operation S210, based on the input information, multiple publishing objects that match the input information are determined from multiple candidate publishing objects.

[0038] In operation S220, the initial recommendation parameters of each of the multiple publishing objects are updated to obtain the target recommendation parameters of each of the multiple publishing objects. The initial recommendation parameters represent the recommendation level of the publishing object.

[0039] In operation S230, the target recommendation parameters and input information of each of the multiple publishing objects are used to filter the content to be recommended for each of the multiple publishing objects, and the target content to be recommended is determined.

[0040] The input information can be search information entered by the user through a terminal device, including but not limited to words, sentences, paragraphs, etc.

[0041] Candidate publishing objects can be a predefined set of objects, an existing set of objects in a database, or any other set of objects that may have an attribution relationship to the content being recommended. Candidate publishing objects can be used for subsequent filtering and matching.

[0042] In some embodiments, candidate publishers can be product sellers, advertisers, video or article publishers, etc. By filtering from multiple candidate publishers to select the few publishers that best match the user's intent, interests, or needs, the scope of publisher recommendations is narrowed down.

[0043] The recommendation rating of a published object represents the probability that the published object will be recommended to a user. The initial recommendation parameters characterize the recommendation rating of the published object.

[0044] The number of content items to be recommended by a publishing entity is usually multiple. After determining the publishing entity, target content can be identified from the content items to be recommended by the publishing entity based on the input information and target recommendation parameters.

[0045] In some embodiments, business information of the publishing object can be obtained, such as the types of goods the product deals in, its business scope, etc. Based on user input information and business information, target content is determined from the content to be recommended. This disclosure is not limited to this. Alternatively, content identical or similar to the input information can be queried from the content to be recommended published by the publishing object, and used as target content. For example, an advertiser publishes an advertisement matching the input information, or a video or article creator publishes a video or article matching the input information.

[0046] In the relevant examples, the initial recommendation parameters and input information of multiple publishing objects can be used to filter the content to be recommended by each publishing object and determine the target content for recommendation.

[0047] However, it should be understood that the subsequent selection of recommended content considers not only the input information but also the initial recommendation parameters of each publishing entity. Therefore, the higher the initial recommendation parameters of a candidate publishing entity, the greater the likelihood that its content will be recommended to the user. In other words, the content of a publishing entity with higher initial recommendation parameters is more likely to be recommended to the user, while the content of a publishing entity with lower initial recommendation parameters may have a lower recommendation level and may not be recommended to the user. This results in the recommendation results being concentrated on publishing entities with higher initial recommendation parameters, reducing the diversity of the target content for recommendation.

[0048] According to embodiments of this disclosure, the initial recommendation parameters of the publishing object can be downweighted to reduce the influence of the publishing object on the generation of target content and reduce the gap in recommendation levels between different publishing objects, so that the content to be recommended by publishing objects with low to medium recommendation levels may also be recommended to users.

[0049] According to the recommendation method disclosed herein, target recommendation parameters for each of the multiple publishing objects are obtained by updating their respective recommendation levels. Based on the target recommendation parameters and the search content, target content for recommendation is determined. Thus, by adjusting the recommendation levels of the multiple publishing objects, the diversity of subsequent target content for recommendation is improved while ensuring the recommendation success rate.

[0050] According to embodiments of this disclosure, for example, Figure 2 The illustrated operation S210, based on input information, determines multiple publishing objects that match the input information from multiple candidate publishing objects. This may include: generating text using the input information to obtain multiple generated objects; and determining multiple publishing objects from the multiple candidate publishing objects based on the multiple generated objects.

[0051] In some embodiments, input information can be fed into a language model, which then generates multiple generated objects associated with the input information. The language model can be pre-trained using a deep learning algorithm, such as a Transformer (encoder-decoder). Specifically, the input information can be encoded using an encoder to obtain an encoded vector, which is then input into a decoder to generate text, resulting in multiple generated objects. This disclosure is not limited to this; the language model can also be a large language model. Any deep learning model capable of generating text and obtaining multiple generated objects based on input information is acceptable.

[0052] Multiple generated objects associated with the input information can be understood as multiple generated objects that the user is interested in. For example, based on the input information, the content to be recommended in the generated objects can be recommended to the user as target content, increasing the probability that the user will perform an action that interests them.

[0053] After obtaining the generated object, it can be matched with the candidate publication object. When the generated object successfully matches the candidate publication object, it is determined that the generated object is the publication object. If the match fails, the generated object can be discarded.

[0054] According to embodiments of this disclosure, multiple generated objects can be quickly obtained by using input information to generate text. The generated objects are then matched with candidate publishing objects to obtain the publishing object. This ensures that the publishing object is determined from a known set of objects, thereby improving the reliability of the publishing object and increasing the recommendation success rate.

[0055] According to embodiments of this disclosure, when performing such Figure 2 Prior to operation S220, the recommendation method may further include: determining initial recommendation parameters based on the matching probability between the published object and the input information. The matching probability represents the probability of performing a predetermined operation on the content to be recommended for the published object in response to the input information.

[0056] A pre-selected action can be an action performed on the content to be recommended by the publishing object to indicate interest. For example, a pre-selected action can include clicking, saving, or sharing.

[0057] The matching probability can range from 0 to 1. The higher the matching probability between the publishing object and the input information, the greater the probability of obtaining the target content from the content to be recommended by that publishing object.

[0058] The similarity between the published object and the input information can be compared using object attribute information such as traffic, the number of recommended content items, and the content type of the recommended content. However, this is not the only approach. A language model used to generate objects can also be trained to output the matching probability between each generated object and the input information. This matching probability can then be used as the matching probability of the published object that matches the generated object. Any method that can characterize the probability that the input information will perform a predetermined operation on the recommended content of the published object is acceptable.

[0059] In some embodiments, a matching probability threshold can be preset, and generated objects that have a matching probability greater than the matching probability threshold and match the candidate publishing objects can be used as publishing objects.

[0060] In some embodiments, the matching probability can be directly used as the initial recommendation parameter, but it is not limited to this. The initial recommendation parameter can also be obtained by normalizing the matching probability.

[0061] According to embodiments of this disclosure, initial recommendation parameters are determined based on matching probability. By utilizing the characteristic that matching probability represents the probability of performing a predetermined operation on the content to be recommended for the publishing object based on input information, the accuracy and effectiveness of the initial recommendation parameters in representing the recommendation level can be ensured.

[0062] According to embodiments of this disclosure, for example, Figure 2 The operation shown updates the initial recommendation parameters of multiple publishing objects to obtain the target recommendation parameters of each publishing object. This can include: using a diversity parameter to update the initial recommendation parameters of multiple publishing objects to obtain the target recommendation parameters of each publishing object. Here, the target content used for recommendation includes multiple items, and the diversity parameter is used to adjust the diversity of the multiple target contents.

[0063] The diversity parameter can range from 0 to 1; for example, it can be 0.6. The diversity parameter can be used to uniformly reduce the weight of the initial recommendation parameters for each publishing object, resulting in the target recommendation parameter for each publishing object, which is 0.6 * the initial recommendation parameter, thus achieving the update.

[0064] For example, the initial recommended parameters for publication object A, publication object B, and publication object C are 0.5, 0.3, and 0.2, respectively. After updating using diversity parameters, the target recommended parameters for publication object A, publication object B, and publication object C are 0.3, 0.18, and 0.12, respectively.

[0065] This reduces the gap in matching probabilities between different publishing objects. In the process of determining target content, while generating target content by combining the target recommendation parameters of the publishing object, the impact of the target recommendation parameters of the publishing object on the generation of target content is reduced. Target content can be obtained from more publishing objects, thereby improving the diversity of target content.

[0066] According to embodiments of this disclosure, the diversity parameter can be determined by: using multiple candidate diversity parameters to determine the sample target content for recommendation of the sample input information; and determining the diversity parameter from multiple candidate diversity parameters based on the diversity evaluation results of the multiple sample target contents.

[0067] The target content of the sample can be obtained by performing operations S210~S230 based on the sample data. The sample data may include sample input information and sample recommended content tags corresponding to the sample input information. The sample recommended content tags may be recommended content determined based on user actions such as clicking or saving.

[0068] The performance of multiple candidate diversity parameters can be evaluated using sample data, yielding individual performance evaluation results for each parameter. Based on these results, the desired diversity parameter is then determined from among the candidate parameters.

[0069] The performance evaluation results can be determined as follows: Based on the sample input information in the sample data, multiple sample publishing objects are determined from multiple candidate publishing objects. Using the candidate diversity parameter, the recommendation level of each of the multiple sample publishing objects is updated, resulting in the sample target recommendation parameters for each of the multiple sample publishing objects. Using the sample target recommendation parameters and sample input information of each of the multiple sample publishing objects, the content to be recommended for each of the multiple sample publishing objects is filtered to determine the sample target content for recommendation. Based on the sample target content and the sample recommendation content tags in the sample data, the performance evaluation results for the candidate diversity parameter are determined.

[0070] In some embodiments, the evaluation metrics for performance evaluation results include at least one of the following: diversity of recommended content, filtering rate, and recommendation success rate.

[0071] In some embodiments, the diversity of recommended content may include the ratio of the types of target recommended content to the total number of target recommended content. The filtering rate characterizes the speed at which target content is determined from the list of objects to be recommended. The recommendation success rate may include whether the target recommended content includes sample recommended content tags, and the ratio of sample recommended content tags to the total number of target recommended content.

[0072] In some embodiments, the multiple candidate diversity parameters can be a predetermined number of values ​​ranging from 0 to 1.

[0073] It should be understood that the smaller the value of the candidate diversity parameter, the more it can reduce the gap between the recommendation levels of different publishing objects, and the better the diversity of the target content will be. However, if the value of the diversity parameter is too small, it will reduce the efficiency of target content generation and increase the additional inference time cost.

[0074] Therefore, by using the diversity assessment results to determine the diversity parameter from multiple candidate diversity parameters, the obtained diversity parameter can ensure that the recommended content has sufficient diversity, while avoiding the low efficiency of target content generation due to the diversity parameter being too small.

[0075] Figure 3 A flowchart illustrating the determination of the publishing target and target recommendation parameters according to an embodiment of this disclosure is shown schematically.

[0076] like Figure 3 As shown, input information 310 is fed into language model M320 to obtain multiple generated objects 350 and initial recommendation parameters 360 for each generated object 350. The multiple generated objects 350 are matched with multiple candidate publishing objects 330 to obtain multiple publishing objects 370. The initial recommendation parameters 360 are updated using diversity parameters 340 to obtain target recommendation parameters 380 for each of the multiple publishing objects 370.

[0077] Figure 4 A flowchart of a recommended method according to another embodiment of this disclosure is illustrated schematically.

[0078] like Figure 4 As shown, the method includes operations S410~S440.

[0079] In operation S410, based on the input information, multiple publishing objects that match the input information are determined from multiple candidate publishing objects.

[0080] In operation S420, the initial recommendation parameters of multiple publishing objects are updated using diversity parameters to obtain the target recommendation parameters of multiple publishing objects.

[0081] When operating S430, text is generated using input information and multiple publishing objects, resulting in multiple generated contents.

[0082] In operation S440, based on the recommendation parameters of each of the multiple generated content items and the target recommendation parameters of each of the multiple publishing objects, the target content is determined from the content to be recommended by each of the multiple publishing objects. The recommendation parameters are used to characterize the recommendation level of the generated content.

[0083] An encoder can be used to encode the input information and multiple generating objects separately, obtaining encoded vectors for the input information and individual encoded vectors for each of the multiple publishing objects. The combined encoded vectors of the input information and the individual encoded vectors of the multiple publishing objects are then input into a decoder for text generation, resulting in multiple generated content items. This disclosure is not limited to this; the input information and multiple publishing objects can also be input together into a large language model to directly generate multiple generated content items.

[0084] Similarity comparisons between generated content and input information can be used to determine the matching probability between the generated content and the input information. Recommendation parameters for each generated content can then be determined based on these matching probabilities. However, this is not the only approach. A language model used to generate content can also be trained to output the matching probability between each generated content and the input information while generating the content. This matching probability can then be used as the recommendation parameter for the generated content.

[0085] According to embodiments of this disclosure, by utilizing recommendation parameters of generated content and target recommendation parameters of publishing objects, target content can be determined from the content to be recommended by multiple publishing objects, thereby more accurately recommending target content that meets user needs and expectations.

[0086] According to embodiments of this disclosure, when performing such Figure 4 Operation S440 may include: determining candidate recommended content from the recommended content of each of multiple publishing objects based on multiple generated content. Determining target content from the candidate recommended content based on the recommendation parameters of each of the multiple generated content and the target recommendation parameters of each of the multiple publishing objects. Content matching or semantic similarity matching may be performed on the generated content and the recommended content of the publishing objects to obtain a similarity matching result. Based on the similarity matching result, candidate recommended content matching the generated content is determined from the recommended content of the publishing objects. The recommendation parameters of the generated content matching the candidate recommended content are used as the recommendation parameters of the candidate recommended content.

[0087] The target recommendation parameters of the publishing object and the recommendation parameters of the candidate recommended content can be weighted and summed to obtain the content recommendation parameters of the candidate recommended content. Based on the content recommendation parameters of multiple candidate recommended content, the target content is determined from the multiple candidate recommended content.

[0088] In some embodiments, the generated content may include keywords. Based on the generated content, candidate keywords that match the keywords can be queried in the content to be recommended of the publishing object. Candidate recommended content is obtained based on multiple matching candidate keywords.

[0089] In some embodiments, candidate recommended content can be scored based on recommendation parameters of each of the generated content to obtain a first score. Published objects can be scored based on target recommendation parameters of each of the multiple publishing objects to obtain a second score. Target content is determined from the candidate recommended content based on the first and second scores. Alternatively, the recommendation parameters of each of the generated content and the target recommendation parameters of each publishing object can be weighted and summed to obtain recommendation parameters for the candidate recommended content. Target content is determined from the candidate recommended content based on these recommendation parameters.

[0090] In some embodiments, a score threshold can be preset, and candidate recommended content with a score greater than the score threshold is selected as target content. This disclosure is not limited to this. Alternatively, candidate recommended content can be sorted based on the sum of a first score and a second score, and a predetermined number of candidate recommended content can be selected as target content.

[0091] According to embodiments of this disclosure, candidate recommended content is first determined by generating content, and then target content is selected from the candidate recommended content by utilizing the recommendation parameters of each generated content and the target recommendation parameters of each publishing object, thereby improving the accuracy of the target content.

[0092] According to embodiments of this disclosure, when performing such Figure 4 The operation S440 shown may further include: determining candidate content nodes that are associated with object nodes from an index tree based on multiple generated content items. The index tree includes nodes and edges; nodes include object nodes and content nodes; object nodes represent publishing objects; content nodes represent keywords in the content to be recommended; and edges represent the relationships between multiple nodes. The operation further includes determining target content nodes from the candidate content nodes based on the recommendation parameters of each of the multiple generated content items and the target recommendation parameters of each of the multiple publishing objects; and determining target content based on the target content nodes.

[0093] The index tree can include nodes and edges. Nodes include object nodes and content nodes. Object nodes represent publishing objects, content nodes represent keywords in the content to be recommended, and edges represent the relationships between multiple nodes. Based on the relationship between candidate publishing objects and content to be recommended, corresponding content nodes are generated under each object node. Based on the relationships between each keyword, the hierarchical relationship between multiple content nodes is obtained, thus completing the construction of the index tree.

[0094] In some embodiments, the hierarchical relationship of multiple content nodes can be determined based on the order of description of the statements. For example, based on the description order of "what to do about toothache caused by cavities," the hierarchical relationship of multiple keywords such as "cavities," "toothache," "how," and "do" can be determined sequentially. Alternatively, the hierarchical relationship of multiple content nodes can be determined based on the inclusion or logical relationship of multiple keywords. For example, for smartphones priced within range B, the hierarchical relationship of multiple keywords such as "phone," "smart," "price," and "range B" can be determined sequentially.

[0095] Multiple content nodes connected at multiple levels form a content node group. The candidate recommended content corresponds to the candidate content node group of the corresponding publishing object, and the multiple candidate keywords in the candidate recommended content correspond one-to-one with the multiple candidate content nodes of the corresponding publishing object.

[0096] Based on multiple generated content items, candidate content nodes that are associated with object nodes are identified from the index tree. This can be achieved by identifying multiple object nodes that match multiple published objects from the index tree, based on the published objects in the generated content. Furthermore, based on keywords in the generated content, multiple candidate content nodes that match the keywords are identified from the content nodes under multiple object nodes.

[0097] Node hierarchy can be determined based on the relationships between multiple nodes. For example, the object node, as the root node, belongs to the same hierarchy, which is the first level. Content nodes connected to the object node are also at the same hierarchy, which is the second level. This process continues until the final leaf node is reached, resulting in the last level. Candidate content nodes for each hierarchy are determined by summing the recommendation parameters of the current hierarchy, the recommendation parameters of multiple upstream hierarchy levels, and the target recommendation parameters of the published object. This process continues until the final content node level, resulting in multiple candidate content nodes.

[0098] You can select a predetermined number or more than a predetermined score of candidate content nodes from multiple candidate content nodes as the target content node.

[0099] In some embodiments, input information, multiple publishing objects, and multiple candidate keyword groups can be used to determine multiple candidate content nodes at the i-th level from the content nodes at the i-th level in the index tree. The candidate keyword groups correspond to related candidate content node groups in the index tree, which include multiple related candidate content nodes upstream of the i-th level candidate content nodes. Based on the target recommendation parameters of each of the multiple publishing objects, the recommendation parameters corresponding to each of the multiple i-th level candidate content nodes, and the recommendation parameter groups corresponding to each of the multiple candidate keyword groups, multiple candidate keywords at the i-th level are determined. This process is repeated until i equals I, where I is the last level. Based on the multiple i-th level candidate keywords and multiple candidate keyword groups, the content to be recommended for each of the multiple publishing objects is filtered to determine the target content.

[0100] In some embodiments, a beam search algorithm can be used to search the index tree to identify multiple candidate content nodes. Specifically, at each level of the index tree, the content nodes are sorted according to the content recommendation parameters, leaving only a predetermined number of content nodes. The search continues to expand at the next level of these content nodes, while other content nodes are pruned, until the last level is searched, thereby obtaining multiple candidate keyword groups.

[0101] In relevant examples, when using the cluster search algorithm to search the index tree, in order to increase the diversity of recommended content, a diversity parameter is usually introduced at each node to penalize keywords that have already appeared in the previous level, thereby encouraging the generation of different results. This approach modifies diversity on a per-word basis, and cannot combine multiple words as a whole to adjust the diversity of recommended content. Furthermore, each word element basically relies on manual parameter tuning, which is computationally intensive and affects processing efficiency.

[0102] According to the embodiments of this disclosure, when searching the index tree, since only the matching probability of the published object node is changed, such as updating the initial recommendation parameter to the target recommendation parameter, more object nodes can be indexed, without changing the matching probability of other content nodes, the diversity of recommended content is guaranteed, while the computational cost in the reasoning process is guaranteed to a certain extent, and the processing efficiency remains basically unchanged.

[0103] Figure 5 The schematic diagram illustrates the principle of determining target content based on an index tree according to an embodiment of the present disclosure.

[0104] like Figure 5As shown, based on input information 501, such as "what to do about toothache caused by cavities," multiple publishing objects 502 and their corresponding initial recommendation parameters 507 are obtained from multiple object nodes in the index tree, such as the first object node 513, the second object node, the third object node, and the fourth object node. The initial recommendation parameters 507 are then weighted using diversity parameters to obtain target recommendation parameters 508. Based on the target recommendation parameters 508, multiple target object nodes are determined from the multiple object nodes. Finally, multiple publishing objects are determined from the multiple candidate publishing objects using the multiple target object nodes. Based on input information 501 and multiple publishing objects, a beam search algorithm is used to search content nodes at each level under multiple target object nodes, resulting in multiple candidate keyword groups and corresponding recommendation parameter groups. Each candidate keyword group can include multiple keywords, such as the first candidate keyword 514 (e.g., "tooth decay"); the second candidate keyword 515 (e.g., "toothache"); the third candidate keyword 516 (e.g., "how"); and the fourth candidate keyword 517 (e.g., "do"). Each recommendation parameter group includes multiple recommendation parameters, such as the first recommendation parameter 509, the second recommendation parameter 510, the third recommendation parameter 511, and the fourth recommendation parameter 512. Based on the target recommendation parameter 508, the first recommendation parameter 509, the second recommendation parameter 510, the third recommendation parameter 511, and the fourth recommendation parameter 512, a score 518 is obtained for each candidate keyword group. Based on the score 518 of each candidate keyword group, the target content is determined from the multiple candidate keyword groups.

[0105] Figure 6 A block diagram of a recommended apparatus according to an embodiment of the present disclosure is shown schematically.

[0106] like Figure 6 As shown, the recommendation device includes a first determination module 610, an update module 620, and a second determination module 630.

[0107] The first determining module 610 is used to determine, based on input information, multiple publishing objects that match the input information from multiple candidate publishing objects.

[0108] The update module 620 is used to update the initial recommendation parameters of multiple publishing objects to obtain the target recommendation parameters of multiple publishing objects, wherein the initial recommendation parameters represent the recommendation level of the publishing object.

[0109] The second determining module 630 is used to filter the content to be recommended by each of the multiple publishing objects using their respective target recommendation parameters and input information, and to determine the target content for recommendation.

[0110] According to embodiments of this disclosure, update module 620 includes update submodule.

[0111] The update submodule is used to update the initial recommendation parameters of multiple publishing objects using the diversity parameter, thereby obtaining the target recommendation parameters of multiple publishing objects. The target content used for recommendation includes multiple items, and the diversity parameter is used to adjust the diversity of multiple target contents.

[0112] According to embodiments of this disclosure, the diversity parameter is determined by the following unit:

[0113] The sample target content unit is used to determine the sample target content for recommendation based on the sample input information using multiple candidate diversity parameters.

[0114] The diversity parameter unit is used to determine the diversity parameter from multiple candidate diversity parameters based on the diversity assessment results of the target content of multiple samples.

[0115] According to embodiments of this disclosure, the second determining module 630 includes a text generation submodule and a target content determining submodule.

[0116] The text generation submodule is used to generate text using input information and multiple publishing objects, resulting in multiple generated contents.

[0117] The target content determination submodule is used to determine the target content from the content to be recommended by multiple publishing objects based on the recommendation parameters of each of the generated content and the target recommendation parameters of each of the multiple publishing objects. The recommendation parameters are used to characterize the recommendation level of the generated content.

[0118] According to embodiments of this disclosure, the target content determination submodule includes a first determination unit and a second determination unit.

[0119] The first determining unit is used to determine candidate recommended content from the recommended content of each of the multiple publishing objects based on multiple generated content.

[0120] The second determining unit is used to determine the target content from the candidate recommended content based on the recommendation parameters of each of the multiple generated content and the target recommendation parameters of each of the multiple publishing objects.

[0121] According to embodiments of this disclosure, the target content determination submodule includes a third determination unit, a fourth determination unit, and a fifth determination unit.

[0122] The third determining unit is used to determine candidate content nodes that are associated with object nodes from the index tree based on multiple generated content. The index tree includes nodes and edges. Nodes include object nodes and content nodes. Object nodes represent publishing objects, content nodes represent keywords in the content to be recommended, and edges represent the association relationships between multiple nodes.

[0123] The fourth determining unit is used to determine the target content node from the candidate content nodes based on the recommendation parameters of each of the multiple generated content nodes and the target recommendation parameters of each of the multiple publishing objects.

[0124] The fifth determination unit determines the target content based on the target content node.

[0125] According to embodiments of this disclosure, the first determining module 610 includes a generation submodule and an object determining module.

[0126] The generation submodule is used to generate text from the input information, resulting in multiple generated objects.

[0127] The object determination submodule is used to determine multiple release objects from multiple candidate release objects based on multiple generated objects.

[0128] According to embodiments of this disclosure, a third determining module is also included.

[0129] The third determining module is used to determine the initial recommendation parameters based on the matching probability between the publishing object and the input information, wherein the matching probability represents the probability of performing a predetermined operation on the content to be recommended by the publishing object for the input information.

[0130] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0131] According to an embodiment of the present disclosure, an electronic device includes: 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, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0132] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described above.

[0133] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0134] Figure 7A schematic block diagram of an example electronic device 700 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.

[0135] like Figure 7 As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded into random access memory (RAM) 703 from storage unit 708. The RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0136] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0137] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 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 701 performs the various methods and processes described above, such as the recommended methods. For example, in some embodiments, the recommended methods may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the recommended methods described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform the recommended methods by any other suitable means (e.g., by means of firmware).

[0138] 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), complex 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.

[0139] 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.

[0140] 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.

[0141] 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).

[0142] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user 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., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0143] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0144] 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.

[0145] 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, comprising: Based on the input information, multiple publishing objects that match the input information are determined from multiple candidate publishing objects; Using diversity parameters, the initial recommendation parameters of each of the multiple publishing objects are updated to obtain the target recommendation parameters of each of the multiple publishing objects, wherein the initial recommendation parameters characterize the recommendation level of the publishing object; and Using the target recommendation parameters of each of the multiple publishing objects and the input information, the content to be recommended by each of the multiple publishing objects is filtered to determine the target content for recommendation; The target content used for recommendation includes multiple items, and the diversity parameter is used to reduce the gap between multiple publishing objects and adjust the diversity of multiple target contents. The diversity parameter is determined in the following way: Multiple candidate diversity parameters are used to determine the target content of the sample input information in the sample data for recommendation. The sample data also includes sample recommendation content tags corresponding to the sample input information. Based on the target content of the sample and the recommended content tags of the sample, determine the diversity assessment results for the candidate diversity parameters; and The diversity parameter is determined from the plurality of candidate diversity parameters based on the diversity assessment results of each of the candidate diversity parameters.

2. The method according to claim 1, wherein, The step of filtering the content to be recommended for each of the multiple publishing objects using their respective target recommendation parameters and the input information, and determining the target content for recommendation, includes: Using the input information and multiple publishing objects, text is generated to obtain multiple generated contents; and Based on the recommendation parameters of each of the generated content and the target recommendation parameters of each of the publishing objects, the target content is determined from the content to be recommended by each of the publishing objects, wherein the recommendation parameters are used to characterize the recommendation level of the generated content.

3. The method according to claim 2, wherein, The step of determining the target content from the content to be recommended by each of the multiple generated content items and the target recommendation parameters of each of the multiple publishing objects includes: Based on the generated content, candidate recommended content is determined from the recommended content of each of the publishing objects; and The target content is determined from the candidate recommended content based on the recommendation parameters of each of the multiple generated content items and the target recommendation parameters of each of the multiple publishing objects.

4. The method according to claim 2, wherein, The step of determining the target content from the content to be recommended by each of the multiple generated content items and the target recommendation parameters of each of the multiple publishing objects includes: Based on multiple generated contents, candidate content nodes that are associated with object nodes are determined from an index tree. The index tree includes nodes and edges. The nodes include object nodes and content nodes. The object node represents the publishing object. The content node represents the keywords in the content to be recommended. The edges represent the association relationships between multiple nodes. Based on the recommendation parameters of each of the generated content items and the target recommendation parameters of each of the publishing objects, a target content node is determined from the candidate content nodes; and The target content is determined based on the target content node.

5. The method according to any one of claims 1 to 4, wherein, The step of determining multiple publishing objects that match the input information from multiple candidate publishing objects based on the input information includes: Using the input information, text is generated to obtain multiple generated objects; and Based on the multiple generated objects, the multiple release objects are determined from the multiple candidate release objects.

6. The method according to any one of claims 1 to 4, further comprising: The initial recommendation parameters are determined based on the matching probability between the published object and the input information, wherein the matching probability represents the probability of performing a predetermined operation on the content to be recommended by the published object in response to the input information.

7. A recommended device, comprising: The first determining module is used to determine, based on input information, multiple publishing objects that match the input information from multiple candidate publishing objects; An update module is used to update the initial recommendation parameters of each of the multiple publishing objects to obtain the target recommendation parameters of each of the multiple publishing objects, wherein the initial recommendation parameters characterize the recommendation level of the publishing object; and The second determining module is used to filter the content to be recommended for each of the multiple publishing objects using their respective target recommendation parameters and the input information, and to determine the target content for recommendation. The update module includes: An update submodule is used to update the initial recommendation parameters of each of the multiple publishing objects using diversity parameters, thereby obtaining the target recommendation parameters of each of the multiple publishing objects. The target content used for recommendation includes multiple items, and the diversity parameters are used to reduce the gap between the multiple publishing objects and adjust the diversity of the multiple target contents. The diversity parameters are determined as follows: Multiple candidate diversity parameters are used to determine the target content of the sample input information in the sample data for recommendation. The sample data also includes sample recommendation content tags corresponding to the sample input information. Based on the target content of the sample and the recommended content tags of the sample, determine the diversity assessment results for the candidate diversity parameters; and The diversity parameter is determined from the plurality of candidate diversity parameters based on the diversity assessment results of each of the candidate diversity parameters.

8. The apparatus according to claim 7, wherein, The second determining module includes: The text generation submodule is used to generate text using the input information and multiple publishing objects to obtain multiple generated contents; and The target content determination submodule is used to determine the target content from the content to be recommended by each of the multiple generated content and the target recommendation parameters of each of the multiple publishing objects, wherein the recommendation parameters are used to characterize the recommendation level of the generated content.

9. The apparatus according to claim 8, wherein, The target content determination submodule includes: The first determining unit is configured to determine candidate recommended content from the recommended content of each of the multiple published objects based on the multiple generated content; and The second determining unit is used to determine the target content from the candidate recommended content based on the recommendation parameters of each of the multiple generated content and the target recommendation parameters of each of the multiple publishing objects.

10. The apparatus according to claim 8, wherein, The target content determination submodule includes: The third determining unit is used to determine, based on multiple generated contents, candidate content nodes that are associated with object nodes from an index tree, wherein the index tree includes nodes and edges, the nodes include object nodes and content nodes, the object nodes represent the publishing object, the content nodes represent keywords in the content to be recommended, and the edges represent the association relationships between multiple nodes; The fourth determining unit is configured to determine a target content node from the candidate content nodes based on the recommendation parameters of each of the multiple generated content nodes and the target recommendation parameters of each of the multiple publishing objects; and The fifth determining unit determines the target content based on the target content node.

11. The apparatus according to any one of claims 7 to 10, wherein, The first determination module includes: A generation submodule is used to generate text using the input information, resulting in multiple generated objects; and An object determination module is used to determine the plurality of publication objects from the plurality of candidate publication objects based on the plurality of generated objects.

12. The apparatus according to any one of claims 7 to 10, further comprising: The third determining module is used to determine the initial recommendation parameters based on the matching probability between the publishing object and the input information, wherein the matching probability represents the probability of performing a predetermined operation on the content to be recommended by the publishing object in response to the input information.

13. 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 1-6.

14. 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 1-6.

15. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images