Article recommendation method, system, device and equipment, storage medium and program product

Abstract

The invention provides an article recommendation method, system, device and equipment, a storage medium and a program product, which can be applied to the technical field of big data. The article recommendation method comprises the following steps: calling a data receiving engine to obtain real-time streaming data of a target object interacting with an article through a terminal from a message queue; processing the real-time streaming data by using a state engine to obtain historical exposure data of the cold start article, the interaction degree of the cold start article and the target object being less than a predetermined value; feature calculation is conducted on the historical exposure data, the sorting score of the cold start article and a content score obtained according to the article attribute of the cold start article, a target recommendation value of the cold start article is obtained, and the sorting score is obtained by processing the historical exposure data through a sorting model; and sending the target recommendation value to a strategy engine, so that the strategy engine configures a recommendation strategy for the cold start article according to the target recommendation value.

Description

Technical Field

[0001] This disclosure relates to the field of big data technology, and more specifically, to a method, system, apparatus, device, storage medium, and program product for recommending items. Background Technology

[0002] With the development of big data and computer technology, more and more product information is being recommended to users via the internet for browsing. Current methods for recommending products rely on extracting product features, such as matching product features with features of items the user prefers, or relying on the similarity between the user and the product. Feature-based recommendation schemes lack real-time feedback and have poor timeliness; schemes relying on user-item similarity face data sparsity problems, making it difficult to accurately calculate similarity. These issues result in current product recommendation systems exhibiting low recommendation diversity and a poor user experience. Summary of the Invention

[0003] In view of this, the present disclosure provides a method, system, apparatus, device, storage medium, and program product for improving the diversity of item recommendations and user experience.

[0004] One aspect of this disclosure provides an item recommendation method, comprising: invoking a data receiving engine to obtain real-time streaming data of a target object interacting with an item via a terminal from a message queue; processing the real-time streaming data using a state engine to obtain historical exposure data of a cold-start item, wherein the interaction level between the cold-start item and the target object is less than a predetermined value; performing feature calculations on the historical exposure data, the ranking score of the cold-start item, and the content score obtained based on the item attributes of the cold-start item to obtain a target recommendation value for the cold-start item, wherein the ranking score is obtained by processing the historical exposure data using a ranking model; and sending the target recommendation value to a strategy engine so that the strategy engine configures a recommendation strategy for the cold-start item based on the target recommendation value.

[0005] According to embodiments of this disclosure, historical exposure data includes the number of times a cold-start item is exposed within a predetermined time period, and the total number of times multiple items are exposed within the predetermined time period. Feature calculations are performed on the historical exposure data, the ranking score of the cold-start item, and the content score obtained based on the item attributes of the cold-start item to obtain a target recommendation value for the cold-start item. This includes: adjusting the exposure ratio of the total number of exposures to the number of exposures of the cold-start item using a first recommendation coefficient to obtain a first recommendation value, wherein the first recommendation coefficient is used to adjust the influence of the historical exposure data of the cold-start item on the target recommendation value; adjusting the content score using a second recommendation coefficient to obtain a second recommendation value, wherein the second recommendation coefficient is used to adjust the influence of the item attributes of the cold-start item on the target recommendation value; and outputting the target recommendation value based on the ranking score, the first recommendation value, and the second recommendation value.

[0006] According to embodiments of this disclosure, the recommendation strategy includes exposure priority and exposure amount; the method further includes: arranging cold start items according to target recommendation values ​​to obtain an arrangement result; and determining the exposure priority and exposure amount based on the distribution position of the cold start items in the arrangement result.

[0007] According to embodiments of this disclosure, the method further includes: in response to a cold-start item being recommended to a target object, real-time monitoring of the interaction between the target object and the cold-start item to obtain feedback data between the target object and the cold-start item; sending the feedback data to a message queue so that the message queue processes the feedback data into feedback stream data; updating the historical exposure data and ranking score of the cold-start item based on the feedback stream data, and adjusting the recommendation strategy based on the updated historical exposure data and ranking score and the content score of the cold-start item.

[0008] According to embodiments of this disclosure, the method further includes: calling a filtering engine to validate the format of the real-time streaming data using regular expressions to filter out invalid data in the real-time streaming data and obtain valid streaming data; and using the filtering engine to filter out abnormal interaction data in the valid streaming data that exceeds a predetermined interaction frequency to obtain target streaming data.

[0009] According to embodiments of this disclosure, the method further includes: embedding the cold-start items to be recommended into the recommendation list of the target object based on a recommendation strategy; and recommending them to the target object through a terminal based on the recommendation list.

[0010] Another aspect of this disclosure provides an item recommendation system, comprising: a data receiving engine for acquiring real-time streaming data of a target object interacting with an item via a terminal from a message queue; a state engine for processing the real-time streaming data to obtain historical exposure data of cold-start items, wherein the interaction level between the cold-start items and the target object is less than a predetermined value; a feature calculation engine for performing feature calculations on the historical exposure data, the ranking score of the cold-start items, and the content score obtained based on the item attributes of the cold-start items to obtain a target recommendation value for the cold-start items, wherein the ranking score is obtained by processing the historical exposure data using a ranking model; and a strategy engine for receiving the target recommendation value sent by the feature calculation engine, configuring a recommendation strategy for the cold-start items based on the target recommendation value.

[0011] One aspect of this disclosure also provides an item recommendation device, comprising: an acquisition module, configured to invoke a data receiving engine to acquire real-time streaming data of a target object interacting with an item via a terminal from a message queue; a processing module, configured to process the real-time streaming data using a state engine to obtain historical exposure data of a cold-start item, wherein the interaction level between the cold-start item and the target object is less than a predetermined value; a calculation module, configured to perform feature calculations on the historical exposure data, the ranking score of the cold-start item, and the content score obtained based on the item attributes of the cold-start item to obtain a target recommendation value for the cold-start item, wherein the ranking score is obtained by processing the historical exposure data using a ranking model; and a sending module, configured to send the target recommendation value to a strategy engine so that the strategy engine can configure a recommendation strategy for the cold-start item based on the target recommendation value.

[0012] Another aspect of this disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the above-described item recommendation method.

[0013] Another aspect of this disclosure provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to implement the above-described item recommendation method.

[0014] Another aspect of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the above-described item recommendation method.

[0015] According to embodiments of this disclosure, real-time streaming data is acquired; the real-time streaming data is processed to obtain historical exposure data of cold-start items; and a target recommendation value is calculated based on the historical exposure data, the ranking score of the cold-start items, and the content feature score, so that the strategy engine can configure recommendation strategies for cold-start items based on the target recommendation value. Because real-time streaming data is acquired during the item recommendation process, the lack of real-time feedback and poor timeliness of related technologies are at least partially overcome. Because multiple factors such as historical exposure data, ranking score, and content feature score of cold-start items are considered, recommendations are no longer based solely on exposure data, thus at least partially overcoming the data sparsity and difficulty in accurately calculating similarity problems of related technologies. This allows even cold-start items with low exposure but high content scores to have a chance to be recommended to the target audience, improving the diversity of item recommendations and the user experience. Attached Figure Description

[0016] The above and other objects, features and advantages of this disclosure will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0017] Figure 1 An exemplary system architecture to which an item recommendation method can be applied according to embodiments of this disclosure is illustrated;

[0018] Figure 2 A flowchart illustrating an article recommendation method according to an embodiment of the present disclosure is shown schematically;

[0019] Figure 3 An architectural diagram of a recommendation system according to an embodiment of the present disclosure is illustrated schematically;

[0020] Figure 4 An architectural diagram of a recommendation system according to another embodiment of the present disclosure is illustrated schematically;

[0021] Figure 5 A block diagram of an item recommendation device according to an embodiment of the present disclosure is shown schematically;

[0022] Figure 6 A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Detailed Implementation

[0023] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0024] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0025] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0026] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0027] In the embodiments disclosed herein, the collection, updating, analysis, processing, use, transmission, provision, disclosure, and storage of data (e.g., including but not limited to user personal information) comply with relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. In particular, necessary measures have been taken to prevent unauthorized access to user personal information data and to safeguard user personal information security and network security.

[0028] In the embodiments disclosed herein, user authorization or consent is obtained before acquiring or collecting user personal information.

[0029] Currently, in the field of item recommendation, the cold start phenomenon occurs. This phenomenon arises in recommendation systems where items with low exposure lack sufficient user interaction data, making it difficult to accurately recommend them to suitable users, resulting in insufficient item exposure. Cold start items are those exhibiting this phenomenon. Several approaches can be used to address cold start items: methods based on item content features, recommending items by matching their content features with those of items preferred by the user; collaborative filtering methods, recommending items based on user or item similarity; popularity-based strategies, pushing new items to a large number of users; and directly applying the existing Upper Confidence Bound (UCB) algorithm for item recommendation. The original UCB algorithm is a multi-armed slot machine algorithm used to balance exploration and exploitation, selecting the optimal option by calculating the upper confidence boundary value for each option. The original UCB algorithm's formula comprehensively considers the average reward and uncertainty of options without adapting it to business scenarios.

[0030] While the aforementioned methods address the cold start problem for items to some extent, they also have several drawbacks: content feature-based methods rely heavily on feature extraction quality and lack real-time feedback, resulting in poor timeliness; collaborative filtering faces the challenge of sparse data for new items, making accurate similarity calculation difficult; popularity-based recommendation strategies lack specificity and are prone to ineffective recommendations; and directly using the original UCB algorithm fails to consider business characteristics, leading to poor exposure optimization and an inability to effectively support new items. Therefore, current methods for addressing the cold start problem still struggle to effectively recommend cold-start items, exhibiting issues of recommendation diversity and efficiency, ultimately reducing user experience.

[0031] In view of this, the embodiments disclosed herein aim to address the problems of low recommendation diversity, recommendation efficiency, and user experience caused by the cold start phenomenon of items in recommendation scenarios, and overcome the shortcomings of related technologies such as lack of specificity and timeliness, and data sparsity. By combining a real-time stream processing engine (such as Apache Flink, an open-source stream processing framework with high throughput and low latency, capable of efficiently processing large-scale real-time stream data) with the modified UCB algorithm, large-scale real-time stream data can be processed efficiently, improving the exposure efficiency of items and optimizing the user experience.

[0032] Figure 1 An exemplary system architecture 100 to which an item recommendation method can be applied according to embodiments of this disclosure is illustrated. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0033] like Figure 1 As shown, the system architecture 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing communication links between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0034] Users can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 via the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, and / or social media platform software, etc. (for example only).

[0035] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be various electronic devices with displays and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0036] Server 105 can be a server that provides various services, such as a backend management server that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). 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.

[0037] In one embodiment, the interaction data between a user and the item to be recommended via any of the three terminal devices (first terminal device 101, second terminal device 102, and third terminal device 103) can be sent via network 104 to server 105 or other devices or systems that can communicate with server 105. Real-time streaming data can be obtained based on this interaction data. Server 105 can determine the historical exposure data and ranking score of the cold-start item based on the real-time streaming data; generate a target recommendation value based on the ranking score, historical exposure data, and content score of the cold-start item; configure a recommendation strategy for the cold-start item based on the target recommendation value; and recommend the item to be recommended via any of the three terminal devices (first terminal device 101, second terminal device 102, and third terminal device 103) based on the adjusted recommendation strategy.

[0038] It should be noted that the item recommendation method provided in this embodiment can generally be executed by server 105. Correspondingly, the item recommendation system provided in this embodiment can generally be set up in server 105. The item recommendation method provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Correspondingly, the item recommendation system provided in this embodiment can also be set up in a server or server cluster that is different from server 105 and capable of communicating with the first terminal device 101, the second terminal device 102, the third terminal device 103, and / or server 105. Alternatively, the item recommendation method provided in this embodiment can also be executed by the first terminal device 101, the second terminal device 102, or the third terminal device 103, or it can be executed by other terminal devices different from the first terminal device 101, the second terminal device 102, or the third terminal device 103. Accordingly, the item recommendation system provided in this embodiment can also be set in the first terminal device 101, the second terminal device 102 or the third terminal device 103, or in other terminal devices different from the first terminal device 101, the second terminal device 102 or the third terminal device 103.

[0039] For example, the real-time streaming data of the target object and the items to be recommended can be originally stored in any one of the first terminal device 101, the second terminal device 102, or the third terminal device 103 (e.g., the first terminal device 101, but not limited thereto), or stored on an external storage device and can be imported into the first terminal device 101. Then, the first terminal device 101 can execute the item recommendation method provided in the embodiments of this disclosure locally, or send the real-time streaming data to other terminal devices, servers, or server clusters, and have the other terminal devices, servers, or server clusters that receive the real-time streaming data execute the item recommendation method provided in the embodiments of this disclosure.

[0040] 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.

[0041] Figure 2 A flowchart illustrating an article recommendation method according to an embodiment of the present disclosure is shown schematically.

[0042] like Figure 2 As shown, the method includes operations S210 to S240.

[0043] In operation S210, the data receiving engine is invoked to retrieve real-time streaming data of the target object interacting with the item through the terminal from the message queue.

[0044] In operation S220, the state engine is used to process the real-time stream data to obtain the historical exposure data of the cold start item, in which the interaction degree between the cold start item and the target object is less than a predetermined value.

[0045] In operation S230, feature calculations are performed on historical exposure data, the ranking score of cold start items, and the content score obtained based on the item attributes of cold start items to obtain the target recommendation value of cold start items. The ranking score is obtained by processing historical exposure data using a ranking model.

[0046] In operation S240, the target recommendation value is sent to the strategy engine so that the strategy engine can configure a recommendation strategy for cold start items based on the target recommendation value.

[0047] In some embodiments, the target object can be an end user receiving recommendation results, and the item can refer to an item (e.g., an item) that needs to be recommended to the target object. Items can include mainstream items with an interaction level greater than or equal to a predetermined value with the target object, and cold-start items with an interaction level less than a predetermined value. Cold-start items can include items that have never been exposed before; the interaction data of these items can be obtained using the interaction data of similar mainstream items. Exposure can refer to the process of an item being recommended to the target object and being seen by the target object, while exposure efficiency refers to the degree to which an item is effectively recommended to the target object.

[0048] In some embodiments, real-time streaming data can be obtained by using Apache Flink's real-time stream processing capabilities. This involves establishing a connection with a message queue (a data structure used to pass messages between different systems or modules, enabling asynchronous communication, decoupling senders and receivers, and improving system stability and flexibility, such as Kafka) to subscribe to and acquire unprocessed real-time streaming data between users and items. This real-time streaming data can be unstructured logs, including the most original interaction records between users and items, such as raw messages of events like user exposure, clicks, conversations, plays, likes, and follows. The collected real-time streaming data can be anonymized using technical means, such as retaining only user type identifiers, item identifiers, interaction behavior types, and timestamps, to ensure that specific target objects cannot be re-identified based on the real-time streaming data.

[0049] In some embodiments, the data receiving engine may include a Flink Kafka Consumer (Kafka source connector) and an addSource (data source registration) mechanism. A connection to Kafka can be established using the Flink Kafka Consumer, a Kafka data source can be added using the addSource operator, and consumer properties such as bootstrap.servers, group.id, and auto.offset.reset can be configured. Data serialization or deserialization is performed using SimpleStringSchema (string deserializer), and parallelism and checkpointing mechanisms are set to ensure data consistency. For example, based on predefined consumer configuration properties, the Kafka source connector is initialized; operators are registered through the data source, integrating the connector and string deserializer into the stream processing job as the data input point; parallelism is configured for the data source registration operators to improve data ingestion throughput; and a distributed checkpointing mechanism is enabled to persist the state and ensure consistency during data processing.

[0050] In some embodiments, historical exposure data can describe macro-statistics about an item being recommended to a user by the recommendation system within a predetermined time period. For example, historical exposure data may include historical exposure counts and total exposure counts. Historical exposure counts may refer to the total number of times an item was recommended to a user within the predetermined time period; total exposure counts may refer to the total number of times all items were recommended by the recommendation system within the predetermined time period. This exposure data can serve as a basis for calculating the exploratory value of cold-start items.

[0051] In some embodiments, a state engine can be used to process real-time streaming data. The state engine can include a ValueState engine and a MapState engine. ValueState can be used to store the number of exposures and cumulative rewards for each item, while MapState can be used to maintain the user-item interaction history and configure TTL (Time-To-Live) policies to manage the state lifecycle.

[0052] By using the ValueState and MapState engines to process real-time streaming data, historical exposure data of cold-start items can be obtained. For example, based on the user-item interaction history stored in the MapState engine, cold-start items with an interaction level less than a predetermined value can be identified. Based on these cold-start items, the exposure count of the cold-start items within a predetermined time period and the total exposure count of multiple items within the predetermined time period can be determined from the ValueState engine.

[0053] In some embodiments, the ranking score of a cold-start item can refer to the probability that a target object will generate a positive interaction (such as a click) with any item in the future, estimated by a machine learning-based ranking model based on the target object's attributes, item features, and target interaction data. The ranking score of a cold-start item represents the ranking model's quantitative judgment of the user's personalized interests, rather than the user's historical behavior statistics.

[0054] In some embodiments, an item's content score can be obtained based on its attributes. Item attributes refer to the quality of the item itself, an objective measure of its intrinsic value, unrelated to user behavior and determined by the item's content. Examples include whether the content is complete or aesthetically pleasing. Paying attention to item attributes in determining the target recommendation value ensures that the recommendation system considers not only user preferences but also the quality of the item's content, thus improving recommendation accuracy.

[0055] In some embodiments, the target recommendation value can be a comprehensive score that determines the item's exposure priority and exposure volume. By integrating the three elements of ranking score, historical exposure data, and content score, it balances the three major goals of personalized recommendation, new product exploration, and content quality, and serves as the direct basis for the subsequent strategy engine to configure recommendation strategies.

[0056] A recommendation strategy refers to the specific action plan formulated by a recommendation system based on a target recommendation value, used to execute recommendations. A recommendation strategy can include two adjustable dimensions: exposure priority and exposure volume. Exposure priority determines the ranking of an item in a user's recommendation list; a higher priority results in a higher ranking. Exposure volume refers to the planned number of times the item will be displayed within a future timeframe. Greater exposure volume increases the likelihood of the item being seen by the user. By adjusting these two dimensions, the recommendation system can accurately allocate traffic resources to the most promising items for recommendation.

[0057] In some embodiments, cold-start items with an interaction level less than a predetermined value may include items that have never been exposed before. For these unexposed items, when calculating the target recommendation value, mainstream items similar to the unexposed item can be queried from the mainstream items pool. The interaction data of these similar mainstream items is then used to determine the target recommendation value for the unexposed item. After the unexposed item has exposure data, the target recommendation value is updated and optimized to improve the item recommendation process. That is, in the embodiments of this disclosure, exposed items can be continuously optimized based on feedback data, while unexposed items can first have an initial target recommendation value determined, and then be continuously optimized based on feedback data.

[0058] According to embodiments of this disclosure, real-time streaming data is acquired; the real-time streaming data is processed to obtain historical exposure data of cold-start items; and a target recommendation value is calculated based on the historical exposure data, the ranking score of the cold-start items, and the content feature score, so that the strategy engine can configure recommendation strategies for cold-start items based on the target recommendation value. Because real-time streaming data is acquired during the item recommendation process, the lack of real-time feedback and poor timeliness of related technologies are at least partially overcome. Because multiple factors such as historical exposure data, ranking score, and content feature score of cold-start items are considered, recommendations are no longer based solely on exposure data, thus at least partially overcoming the data sparsity and difficulty in accurately calculating similarity problems of related technologies. This allows even cold-start items with low exposure but high content scores to have a chance to be recommended to the target audience, improving the diversity of item recommendations and the user experience.

[0059] In some embodiments, real-time streaming data is processed to remove invalid and abnormal data, resulting in target streaming data. Using this target streaming data to determine target recommendation values ​​can improve the accuracy of determining historical exposure data and predicting interest levels, thereby increasing the accuracy of target recommendation values ​​and item recommendations. Based on this, the method may further include the following operations: calling a filtering engine to validate the format of the real-time streaming data using regular expressions to filter out invalid data and obtain valid streaming data; and using the filtering engine to filter out abnormal interaction data exceeding a predetermined interaction frequency from the valid streaming data to obtain target streaming data.

[0060] In some embodiments, the target stream data can be interactive data obtained after processing real-time stream data through filtering and standardization. This processing removes invalid, erroneous, and abnormal data from the real-time stream data, retaining only reliable interactive data that can be used to calculate the target recommendation value.

[0061] In processing real-time streaming data, the filter engine can be used to filter out invalid and abnormal data, the map operator can be used for data format conversion and standardization, and the flatMap operator can be used to handle the unfolding of complex data structures. Specifically, regular expressions are used to validate the data format and complete the data validity check. User behavior frequency is counted based on a sliding window, and a user is marked as abnormal if the number of clicks on a single piece of content within 5 minutes exceeds 10. The keyBy() function is used to group by user identifier to complete the detection of abnormal data. For example, the filter engine can be used to filter out invalid data that does not conform to the regular expression, resulting in valid streaming data; then, the filter engine can be used again to filter out the abnormal data marked as "user, single content, clicks within 5 minutes > 10" from the valid streaming data, resulting in the target streaming data.

[0062] According to embodiments of this disclosure, by filtering and other processing of real-time streaming data, the waste of computing resources caused by invalid data in the real-time streaming data and the low accuracy of target recommendation value calculation can be avoided, thereby improving the efficiency and accuracy of item recommendation.

[0063] In some embodiments, historical exposure data may include the number of times a cold-start item is exposed within a predetermined time period, and the total number of times multiple items are exposed within the predetermined time period. The process of performing feature calculations on the historical exposure data, the ranking score of the cold-start item, and the content score obtained based on the item attributes of the cold-start item to obtain the target recommendation value of the cold-start item may include the following operations: using a first recommendation coefficient to adjust the exposure ratio of the total number of exposures to the number of exposures of the cold-start item, to obtain a first recommendation value, wherein the first recommendation coefficient is used to adjust the influence of the historical exposure data of the cold-start item on the target recommendation value; using a second recommendation coefficient to adjust the content score, to obtain a second recommendation value, wherein the second recommendation coefficient is used to adjust the influence of the item attributes of the cold-start item on the target recommendation value; and outputting the target recommendation value based on the ranking score, the first recommendation value, and the second recommendation value.

[0064] In some embodiments, the ranking score can be obtained by a ranking model, which can be a pre-trained machine learning model, such as a convolutional neural network. The input of the model can be the interaction data between the user and the item and the user's characteristics. The model can output a ranking score, which can be a quantification of the degree of interest of the user in the item in the future (based on the user's click on the item).

[0065] In some embodiments, the process of obtaining the target recommendation value based on historical exposure data, the ranking score of cold start items, and the content score of cold start items can be as shown in formula (1).

[0066] (1)

[0067] in, Let j represent the target recommended value, where j can represent the j-th item, such as a cold start item. For the target recommended value of the j-th item, The ranking score output by the ranking model. This refers to the number of times the j-th item is exposed to the target audience within a predetermined time period, i.e., the historical exposure count. This is the sum of the number of times all items to be recommended are exposed within the predetermined time period, i.e., the total number of exposures. The first recommended value is... The content score is determined by evaluating the quality of an item using predetermined quality assessment standards (such as content completeness, compliance, aesthetic appeal, and production precision). The second recommended value is... As the first recommendation coefficient, This is the second recommendation coefficient.

[0068] The above process can use Flink's ProcessFunction to handle complex business logic, integrate external ranking models to obtain ranking scores and content scores, update the upper bound of the confidence interval in real time, and support dynamic adjustment of the first and second recommendation coefficients.

[0069] According to embodiments of this disclosure, the modified UCB algorithm incorporates business characteristics, namely, personalized business characteristics, by adding a ranking score predicted based on the attributes of the target object and the interaction data between the target object and the item to formula (1); it also incorporates the business characteristics of content quality by adding a content score unrelated to user behavior to formula (1), which allows a high-quality new item to gain exposure opportunities even if its initial click-through rate (i.e., ranking score) is low. This suppresses the exploration space for low-quality content, actively guides the ecosystem towards high-quality development, and improves the diversity of item recommendations. On the other hand, by recommending items based on predicted interest levels, historical exposure data, and the inherent attributes of the items to be recommended, the information cocoon effect caused to users by recommending items solely based on historical exposure data can be avoided, thus improving the efficiency and accuracy of recommendations.

[0070] In some embodiments, the target recommendation value output by the above formula can facilitate the strategy engine in configuring recommendation strategies for cold-start items and recommending items to be recommended using the configured recommendation strategies. Specifically, the recommendation strategy may include exposure priority and exposure amount; the process of configuring the recommendation strategy is as follows: arranging the cold-start items according to the target recommendation value to obtain the arrangement result; determining the exposure priority and exposure amount based on the distribution position of the cold-start items in the arrangement result.

[0071] In some embodiments, cold-start items can be arranged in descending order of target recommendation value to obtain an arrangement result. The exposure priority and exposure amount are determined based on the item's position in the arrangement result. For example, the earlier the item is arranged, the higher the target recommendation value, and the higher the exposure priority and the greater the exposure amount; conversely, the later the item is arranged, the lower the target recommendation value, and the lower the exposure priority and exposure amount, thus completing the configuration of the recommendation strategy.

[0072] In one embodiment, if the target recommendation value for item 1 is 10, the target recommendation value for item 2 is 6, the target recommendation value for item 3 is 8, the target recommendation value for item 4 is 4, the target recommendation value for item 5 is 2, and the target recommendation value for item 6 is 1, then the arrangement result of arranging the items in descending order of target recommendation value is [item 1, item 3, item 2, item 4, item 5, item 6]. For item 1, the exposure priority can be 3 and the exposure amount is 300; for item 3, the exposure priority can be 2 and the exposure amount is 200; and for item 4, the exposure priority can be 1 and the exposure amount is 100.

[0073] According to embodiments of this disclosure, by assigning tiered recommendation priorities and exposure to cold-start items based on target recommendation values, the recommendation system achieves automated and refined dynamic allocation of resources, thereby improving recommendation efficiency and utilization of recommendation data.

[0074] In some embodiments, the target audience group interested in the items can be identified based on real-time data exchanged between the target object and the items; and recommended items can be suggested to the target audience group based on an adjusted recommendation strategy.

[0075] In some embodiments, based on the distribution characteristics of target objects that interact with the item from real-time streaming data, a target object group interested in the item can be identified. Then, in conjunction with the adjusted recommendation strategy described above, the item can be recommended to this target object group. For example, by combining various attributes and interaction behaviors of the target objects, it can be determined that the target object group interested in the recommended item is located in region K. Therefore, the adjusted recommendation strategy described above can be used to prioritize exposing the recommended item on pages where the target object group in region K is concentrated. This avoids invalid exposure and increases the item's visibility.

[0076] In some embodiments, the following operations can be performed during the process of recommending items to the target object based on the above recommendation strategy: embedding the cold start items to be recommended into the recommendation list of the target object based on the recommendation strategy; and recommending the items to the target object through the terminal based on the recommendation list.

[0077] In some embodiments, the information recommendation list can refer to the list of content that has been fully sorted and filtered by the recommendation system and is ultimately displayed to the target user. This is the final output of the recommendation process. The content in the information recommendation list is mainly sorted based on the user's interests and real-time behavior. The information recommendation list can include items from different strategies and algorithmic channels, such as popular content, collaborative filtering results, and precise ranking results. The information stream seen by the target user after opening the application on the terminal device can serve as an information recommendation list, which can include various forms of information to be recommended, such as videos, images and text, and advertisements. Information to be recommended refers to a single content entity that already exists in the information recommendation list and is ready to be recommended to the target user. Each piece of information to be recommended can correspond to a specific item.

[0078] Information about items to be recommended during a cold start can be embedded into the recommendation list, updating the recommendation list for the target object. Embedding can be done intermittently; for example, if the recommendation list includes items 1 through 10 to be recommended, information about item 1 can be embedded between the second and third items in the recommendation list, and information about item 3 can be embedded between the fourth and fifth items in the recommendation list, resulting in an updated recommendation list. The recommendation operation can then be performed using this updated recommendation list.

[0079] According to embodiments of this disclosure, by using intermittent embedding instead of concentrated embedding when embedding information of items to be recommended, the balance between newly embedded cold-start items and the original mainstream items in the information recommendation list is effectively balanced. This preserves the user's mainstream experience while naturally introducing cold-start items, thereby improving the diversity and efficiency of recommendations.

[0080] In some embodiments, in response to a cold-start item being recommended to a target object, the following operations may be performed: real-time monitoring of the interaction between the target object and the cold-start item to obtain feedback data between the target object and the cold-start item; sending the feedback data to a message queue so that the message queue processes the feedback data into feedback stream data; updating the historical exposure data and ranking score of the cold-start item based on the feedback stream data, and adjusting the recommendation strategy based on the updated historical exposure data and ranking score as well as the content score of the cold-start item.

[0081] In some embodiments, after a cold-start item is recommended and reaches the target object, real-time feedback data between the target object and the recommended item can be monitored, such as clicks and comments. This feedback data is then sent to a message queue, which processes it into a feedback stream. Based on this feedback stream data, historical exposure data and ranking scores can be updated. The target recommendation value for the item is then updated based on the updated exposure data and ranking score, and a recommendation strategy is configured based on the updated target recommendation value. This iterative optimization of item recommendations improves the accuracy and effectiveness of the recommendations.

[0082] According to embodiments of this disclosure, by collecting feedback data between the target object and the recommended items and continuously adjusting and optimizing the item recommendation strategy, the system can achieve self-learning and continuous optimization, thereby improving the intelligence level of the recommendation system.

[0083] Figure 3 An architectural diagram of a recommendation system according to an embodiment of the present disclosure is illustrated schematically.

[0084] like Figure 3 As shown, this recommendation system can implement the item recommendation methods described in operations S210 to S240. The recommendation system may include a data acquisition module 310, a data cleaning and filtering module 320, a content feature calculation module 330, an exposure optimization module 340, a recommendation execution module 350, and a feedback collection module 360.

[0085] The data acquisition module 310 can provide real-time streaming data, such as real-time streaming data and feedback data, to the data cleaning and filtering module 320. The data cleaning and filtering module 320 can provide target interaction data to the content feature calculation module 330. The content feature calculation module 330 can provide target recommendation values ​​to the exposure optimization module 340. The exposure optimization module 340 can provide adjusted recommendation strategies to the recommendation execution module 350. The recommendation execution module 350 can provide recommended items to the target object. The feedback collection module 360 ​​can collect feedback interaction data between the target object and the items and send the feedback interaction data back to the data acquisition module 310. The feedback collection module 360 ​​can serve as the optimization basis for the content feature calculation module 330.

[0086] The data acquisition module 310 can utilize Apache Flink's real-time stream processing capabilities to subscribe to and acquire interaction data between the target object and the item in real time by establishing a connection with a message queue (such as Kafka). This interaction data mainly includes the target object's exposure, clicks, and dialogues with the item. The data acquisition module 310 can implement the acquisition operations in S210.

[0087] The data cleaning and filtering module 320 performs multi-dimensional cleaning and filtering operations after receiving data transmitted from the data acquisition module 310. First, it removes obviously invalid data, such as data with incorrect formatting or null values. Second, it filters out abnormal data by setting reasonable thresholds to identify and eliminate data that may be generated by misoperation or malicious behavior, such as a large number of repeated clicks by the same user on the same content within a short period. After cleaning and filtering, high-quality, valid interaction data is obtained, providing a reliable data foundation for subsequent content feature calculations. The data cleaning and filtering module 320 can implement the processing operations in step S210.

[0088] The content feature calculation module 330 can calculate content features based on the cleaned interaction data using the modified UCB algorithm. Content features refer to the information that an item possesses that can describe its attributes and characteristics, such as category, tag, and attribute. In this embodiment, the UCB algorithm has been modified to suit the business characteristics of the recommendation scenario. The modified UCB algorithm introduces ranking model score and content quality score, and introduces coefficient control weights. The specific calculation formula is shown in formula (1). The content feature calculation module 330 can execute operation S220 and operation S230.

[0089] The exposure optimization module 340 can optimize the exposure of items based on the target recommendation value obtained by the content feature calculation module 330. Items with higher target recommendation values ​​can be given higher exposure priority and greater exposure. Simultaneously, by combining the attributes and historical behavior of the target audience, items are accurately recommended to target audience groups more likely to be interested, avoiding ineffective exposure. For example, for new items with high target recommendation values ​​and a high degree of matching with the preferences of target audiences in region K, priority is given to exposure on pages where target audiences in region K are concentrated. The exposure optimization module 340 can execute the adjustment operation of operation S240.

[0090] The recommendation execution module 350, based on the exposure strategy adjusted by the exposure optimization module 340, embeds items into the target object's recommendation list and other locations to complete the recommendation operation. This recommendation execution module 350 can interact with the front-end display system to ensure that recommended items are presented to the target object in a timely and accurate manner. The recommendation execution module 350 can execute the recommendation operation of operation S240.

[0091] The feedback collection module 360 ​​can monitor the interaction behavior of the target object with the recommended items in real time, collect user feedback data such as clicks and comments, and transmit this data to the data acquisition module 310 in real time, forming a data closed loop. By continuously collecting feedback data, the content feature calculation module 330 is provided with the latest data support, enabling the modified UCB algorithm to continuously learn and optimize, thereby improving the accuracy and effectiveness of recommendations.

[0092] According to embodiments of this disclosure, large-scale real-time stream processing of interaction data between target objects and items is implemented based on Apache Flink. The UCB algorithm is specifically modified by introducing predicted interest levels, inherent item attributes, and weight coefficients from the ranking model output, making the UCB algorithm more suitable for the cold start problem in recommendation scenarios. On the other hand, regarding real-time stream processing technology, in addition to Apache Flink, secondary development can be performed based on open-source frameworks, or it can be implemented entirely based on self-developed components, with specific adjustments made according to actual needs.

[0093] Figure 4 An architectural diagram of a recommendation system according to another embodiment of the present disclosure is illustrated schematically.

[0094] like Figure 4 As shown, the recommendation system in this embodiment may include a data receiving engine 410, a state engine 420, a feature calculation engine 430, and a strategy engine 440.

[0095] The data receiving engine 410 is used to obtain real-time streaming data of the target object interacting with the item through the terminal from the message queue.

[0096] The State Engine 420 is used to process real-time streaming data to obtain historical exposure data of cold start items, where the interaction level between the cold start item and the target object is less than a predetermined value.

[0097] The feature calculation engine 430 is used to perform feature calculations on historical exposure data, the ranking score of cold start items, and the content score obtained based on the item attributes of cold start items to obtain the target recommendation value of cold start items. The ranking score is obtained by processing historical exposure data using a ranking model.

[0098] The strategy engine 440, such as the rule parser and executor, can be used to receive the target recommendation value sent by the feature calculation engine, and configure the recommendation strategy for cold start items based on the target recommendation value.

[0099] The recommendation system provided in this disclosure addresses the data timeliness issue through its data receiving engine, ensuring that system decisions are based on the latest user feedback. The state engine and feature calculation engine resolve data sparsity and representation issues, aggregating the sparse and scattered initial interaction data of cold-start items into continuous and quantifiable time-series features, providing a basis for evaluation. The strategy engine addresses strategy execution and personalization issues, ensuring that intelligent decisions are accurately implemented and maximizing recommendation efficiency. This real-time data processing and decision-making pipeline automatically solves the problem of low recommendation efficiency for cold-start items, improving the diversity of item recommendations and user experience.

[0100] Figure 5 A block diagram of an item recommendation device according to an embodiment of the present disclosure is shown schematically.

[0101] like Figure 5 As shown, the item recommendation device 500 includes an acquisition module 510, a processing module 520, a calculation module 530, and a sending module 540.

[0102] The acquisition module 510 is used to call the data receiving engine to obtain real-time streaming data of the target object interacting with the item through the terminal from the message queue.

[0103] The processing module 520 is used to process real-time streaming data using a state engine to obtain historical exposure data of cold start items, wherein the interaction degree between the cold start item and the target object is less than a predetermined value.

[0104] The calculation module 530 is used to perform feature calculations on historical exposure data, the ranking score of cold start items, and the content score obtained based on the item attributes of cold start items to obtain the target recommendation value of cold start items. The ranking score is obtained by processing historical exposure data using a ranking model.

[0105] The sending module 540 is used to send the target recommendation value to the strategy engine so that the strategy engine can configure the recommendation strategy for cold start items based on the target recommendation value.

[0106] In some embodiments, the calculation module 530 may include a first adjustment unit, a second adjustment unit, and an output unit.

[0107] The first adjustment unit is used to adjust the exposure ratio of the total number of exposures to the number of exposures of the cold start item using a first recommendation coefficient to obtain a first recommendation value. The first recommendation coefficient is used to adjust the influence of the historical exposure data of the cold start item on the target recommendation value.

[0108] The second adjustment unit is used to adjust the content score using the second recommendation coefficient to obtain the second recommendation value. The second recommendation coefficient is used to adjust the degree of influence of the cold start item's item attributes on the target recommendation value.

[0109] The output unit is used to output the target recommendation value based on the ranking score, the first recommendation value, and the second recommendation value.

[0110] In some embodiments, the item recommendation device 500 may further include an arrangement module and a determination module.

[0111] The sorting module is used to sort cold-start items according to the target recommended value and obtain the sorting result.

[0112] The determination module is used to determine the exposure priority and exposure amount based on the distribution position of cold-start items in the arrangement results.

[0113] In some embodiments, the item recommendation device 500 may further include a monitoring module, a feedback module, and an update module.

[0114] The monitoring module is used to respond to the cold start item being recommended to the target object, monitor the interaction behavior between the target object and the cold start item in real time, and obtain feedback data between the target object and the cold start item.

[0115] The feedback module is used to send feedback data to the message queue, so that the message queue can process the feedback data into feedback stream data.

[0116] The update module is used to update the historical exposure data and ranking scores of cold-start items based on feedback stream data, and to adjust the recommendation strategy based on the updated historical exposure data and ranking scores, as well as the content scores of cold-start items.

[0117] In some embodiments, the item recommendation device 500 may further include a verification module and a filtering module.

[0118] The validation module calls the filtering engine and uses regular expressions to validate the format of the real-time streaming data in order to filter out invalid data and obtain valid streaming data.

[0119] The filtering module is used to filter out abnormal interaction data that exceeds a predetermined interaction frequency from the valid stream data using the filtering engine, so as to obtain the target stream data.

[0120] In some embodiments, the item recommendation device 500 may further include an embedding module and a recommendation module. The embedding module is used to embed the cold-start item to be recommended into the recommendation list of the target object based on a recommendation strategy. The recommendation module is used to recommend the item to the target object via a terminal based on the recommendation list.

[0121] Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure, or at least part of the functions of any one or more of them, can be implemented in one module. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be implemented by dividing them into multiple modules. Any one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as hardware circuitry, such as a Field-Programmable Gate Array (FPGA), a Programmable Logic Array (PLA), a System-on-Chip, a System-on-a-Substrate, a System-on-Package, an Application-Specific Integrated Circuit (ASIC), or implemented in hardware or firmware by any other reasonable means of integrating or packaging circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, one or more of the modules, submodules, units, and subunits according to embodiments of the present disclosure can be at least partially implemented as computer program modules, which, when run, can perform corresponding functions.

[0122] For example, any plurality of the acquisition module 510, processing module 520, calculation module 530, and transmission module 540 may be combined into one module / unit / subunit, or any one of these modules / units / subunits may be split into multiple modules / units / subunits. Alternatively, at least part of the functionality of one or more of these modules / units / subunits may be combined with at least part of the functionality of other modules / units / subunits and implemented in one module / unit / subunit. According to embodiments of the present disclosure, at least one of the acquisition module 510, processing module 520, calculation module 530, and transmission module 540 may be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the acquisition module 510, processing module 520, calculation module 530, and sending module 540 may be implemented at least partially as a computer program module, which can perform corresponding functions when the computer program module is run.

[0123] It should be noted that the item recommendation device part in the embodiments of this disclosure corresponds to the item recommendation method part in the embodiments of this disclosure. For a detailed description of the item recommendation device part, please refer to the item recommendation method part, which will not be repeated here.

[0124] Figure 6A block diagram of an electronic device suitable for implementing the methods described above, according to embodiments of the present disclosure, is illustrated schematically. Figure 6 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0125] like Figure 6 As shown, an electronic device 600 according to an embodiment of this disclosure includes a processor 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage portion 608 into a random access memory (RAM) 603. The processor 601 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 601 may also include onboard memory for caching purposes. The processor 601 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this disclosure.

[0126] RAM 603 stores various programs and data required for the operation of electronic device 600. Processor 601, ROM 602, and RAM 603 are interconnected via bus 604. Processor 601 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 602 and / or RAM 603. It should be noted that the programs may also be stored in one or more memories other than ROM 602 and RAM 603. Processor 601 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0127] According to embodiments of this disclosure, the electronic device 600 may further include an input / output (I / O) interface 605, which is also connected to a bus 604. The electronic device 600 may also include one or more of the following components connected to the input / output (I / O) interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the input / output (I / O) interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 610 as needed so that computer programs read from it can be installed into the storage section 608 as needed.

[0128] According to embodiments of this disclosure, the method flow according to embodiments of this disclosure can be implemented as a computer software program. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by processor 601, it performs the functions defined in the system of embodiments of this disclosure. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0129] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0130] According to embodiments of this disclosure, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0131] For example, according to embodiments of this disclosure, a computer-readable storage medium may include the ROM 602 and / or RAM 603 described above and / or one or more memories other than ROM 602 and RAM 603.

[0132] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this disclosure. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the item recommendation method provided in the embodiments of this disclosure.

[0133] When the computer program is executed by the processor 601, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0134] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via the communication section 609, and / or installed from the removable medium 611. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0135] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on a user's computing device, partially on a user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0136] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions. Those skilled in the art will understand that the features described in the various embodiments of the present disclosure can be combined and / or combined in various ways, even if such combinations are not explicitly described in the present disclosure. In particular, the features described in the various embodiments of this disclosure may be combined and / or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0137] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for recommending items, characterized in that, The method includes: The data receiving engine is invoked to retrieve real-time streaming data of the target object interacting with the item through the terminal from the message queue; The real-time streaming data is processed using a state engine to obtain historical exposure data of the cold start item, wherein the interaction between the cold start item and the target object is less than a predetermined value; The target recommendation value of the cold start item is obtained by performing feature calculations on the historical exposure data, the ranking score of the cold start item, and the content score obtained based on the item attributes of the cold start item. The ranking score is obtained by processing the historical exposure data using a ranking model. The target recommendation value is sent to the strategy engine so that the strategy engine can configure a recommendation strategy for the cold start item based on the target recommendation value.

2. The method according to claim 1, characterized in that, The historical exposure data includes the number of times the cold start item was exposed within a predetermined time period, and the total number of times multiple items were exposed within the predetermined time period; The step of performing feature calculations on the historical exposure data, the ranking score of the cold start item, and the content score obtained based on the item attributes of the cold start item to obtain the target recommendation value of the cold start item includes: Using a first recommendation coefficient, the exposure ratio of the total number of exposures to the number of exposures of the cold start item is adjusted to obtain a first recommendation value. The first recommendation coefficient is used to adjust the degree of influence of the historical exposure data of the cold start item on the target recommendation value. The content score is adjusted using a second recommendation coefficient to obtain a second recommendation value. The second recommendation coefficient is used to adjust the degree of influence of the cold start item's item attributes on the target recommendation value. Based on the ranking score, the first recommendation value, and the second recommendation value, the target recommendation value is output.

3. The method according to claim 1, characterized in that, The recommendation strategy includes exposure priority and exposure volume; The method further includes: The cold start items are arranged according to the target recommendation value to obtain the arrangement result; The exposure priority and the exposure amount are determined based on the distribution position of the cold-start items in the arrangement result.

4. The method according to claim 1, characterized in that, The method further includes: In response to the cold start item being recommended to the target object, the interaction behavior between the target object and the cold start item is monitored in real time to obtain feedback data between the target object and the cold start item; The feedback data is sent to the message queue so that the message queue processes the feedback data into feedback stream data. The historical exposure data and ranking score of the cold start item are updated based on the feedback stream data, and the recommendation strategy is adjusted based on the updated historical exposure data, ranking score, and content score of the cold start item.

5. The method according to claim 1, characterized in that, The method further includes: The filtering engine is invoked, and regular expressions are used to validate the format of the real-time streaming data in order to filter out invalid data in the real-time streaming data and obtain valid streaming data. Using the filtering engine, abnormal interaction data exceeding a predetermined interaction frequency is filtered out from the valid stream data to obtain the target stream data.

6. The method according to claim 1, characterized in that, The method further includes: Based on the recommendation strategy, the cold start items to be recommended are embedded into the recommendation list of the target object; Based on the recommendation list, recommendations are made to the target object through the terminal.

7. An item recommendation system, characterized in that, The system includes: The data receiving engine is used to obtain real-time streaming data of the target object interacting with the item through the terminal from the message queue; A state engine is used to process the real-time streaming data to obtain historical exposure data of cold start items, wherein the interaction degree between the cold start items and the target object is less than a predetermined value; The feature calculation engine is used to perform feature calculations on the historical exposure data, the ranking score of the cold start item, and the content score obtained based on the item attributes of the cold start item to obtain the target recommendation value of the cold start item. The ranking score is obtained by processing the historical exposure data using a ranking model. A strategy engine is used to receive the target recommendation value sent by the feature calculation engine, and configure a recommendation strategy for the cold start item based on the target recommendation value.

8. An item recommendation device, characterized in that, The device includes: The acquisition module is used to call the data receiving engine to retrieve real-time streaming data of the target object interacting with the item through the terminal from the message queue; The processing module is used to process the real-time streaming data using the state engine to obtain the historical exposure data of the cold start item, wherein the interaction degree between the cold start item and the target object is less than a predetermined value; The calculation module is used to perform feature calculations on the historical exposure data, the ranking score of the cold start item, and the content score obtained based on the item attributes of the cold start item to obtain the target recommendation value of the cold start item. The ranking score is obtained by processing the historical exposure data using a ranking model. The sending module is used to send the target recommendation value to the strategy engine, so that the strategy engine can configure a recommendation strategy for the cold start item based on the target recommendation value.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more programs. The characteristic is that, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 6.

10. A computer-readable storage medium having executable instructions stored thereon, characterized in that, When executed by the processor, this instruction causes the processor to implement the method of any one of claims 1 to 6.

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

Images