A resource recommendation method and a corresponding apparatus

Abstract

The disclosure provides a resource recommendation method and a corresponding device, relates to the technical field of computer application, and particularly relates to the technical field of information flow. The specific implementation scheme is as follows: a target application program to which a to-be-recommended resource belongs and a target category to which the to-be-recommended resource belongs are determined; preference intensity information of the target category under a target channel is acquired, the target channel being a channel from which a target user acquires the target application program; the preference intensity information is used to adjust a ranking score of the to-be-recommended resource; and based on the adjusted ranking score, a resource recommended to the target user is determined. The disclosure can effectively improve the recommendation effect in the cold start stage.

Description

Technical Field

[0001] This disclosure relates to the field of computer application technology, and more particularly to the field of information flow technology. Background Technology

[0002] In the field of internet resource recommendation, recommendation systems typically rely on users' historical behavioral data (such as clicks, browsing, and favorites) to build user interest models, thereby enabling personalized resource recommendations to different users. However, for newly registered users or those who have just started using the resource recommendation service, their historical behavioral data is extremely sparse or even completely missing, making it difficult for the recommendation system to accurately determine their interests and preferences. This phenomenon is known as the "cold start" problem. The quality of recommendations during the cold start phase directly affects the user experience, and consequently, user retention rates.

[0003] Therefore, how to more accurately capture the potential interests and preferences of new users, thereby improving the recommendation effect during the cold start phase, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0004] In view of this, the present disclosure provides a resource recommendation method, apparatus, electronic device, computer-readable storage medium, and computer program product to improve the recommendation effect during the cold start phase.

[0005] According to a first aspect of this disclosure, a resource recommendation method is provided, comprising: Determine the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs; Obtain the preference intensity information of the target category under the target channel, where the target channel is the channel from which the target user obtains the target application; The ranking score of the resource to be recommended is adjusted using the preference intensity information; Based on the adjusted ranking score, resources recommended to the target user are determined.

[0006] According to a second aspect of this disclosure, a resource recommendation apparatus is provided, comprising: The determining unit is configured to determine the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs; The acquisition unit is configured to acquire preference intensity information of the target category under the target channel, wherein the target channel is the channel from which the target user acquires the target application; The adjustment unit is configured to adjust the ranking score of the resource to be recommended using the preference intensity information; The recommendation unit is configured to determine the resources to recommend to the target user based on the adjusted ranking score.

[0007] According to a third 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 that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0008] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0009] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0010] It can be seen that by utilizing prior data such as the channel information from which target users acquire target applications, the preference intensity information of target categories under target channels can be obtained. Based on this, the scores of the resources to be recommended can be adjusted during the ranking stage, so that the recommendation results can match the potential interests and preferences of target users related to channels. This solves the problem of mismatch between recommendation results and user interests caused by the lack of user behavior characteristics during the cold start stage, and improves the recommendation effect.

[0011] 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

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure.

[0013] Figure 1 This is a system architecture diagram applicable to the embodiments of this disclosure.

[0014] Figure 2 A flowchart of a resource recommendation method provided in an embodiment of this disclosure.

[0015] Figure 3a This is a schematic diagram of the process for pre-constructing preference intensity information for "channel-category" provided in an embodiment of this disclosure.

[0016] Figure 3b This is a schematic diagram of the cold start phase for a new user in a media streaming application, provided as an embodiment of this disclosure.

[0017] Figure 4A schematic block diagram of a resource recommendation device provided in an embodiment of this disclosure.

[0018] Figure 5 This is a block diagram of an electronic device used to implement the resource recommendation method of the embodiments of this disclosure. Detailed Implementation

[0019] 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.

[0020] In the field of resource recommendation systems, to achieve personalized recommendations for new users during the cold start phase, related technologies commonly employ methods that build interest models based on users' historical behavioral data. Specifically, this approach collects and analyzes user behavior logs such as clicks, favorites, and shares to create a user behavior profile for each user, incorporating multiple category preference weights. Its basic working principle is that when a user generates new browsing behavior, the system promptly updates the weight distribution in their behavioral features and recalls and ranks candidate content based on these features. Its widespread application stems primarily from its ability to effectively personalize recommendation results based on user feedback.

[0021] However, this approach doesn't perform well when applied to cold-start scenarios where new users are using recommendation services. A fundamental contradiction lies in the fact that, in pursuit of accuracy, this approach relies on historical user behavior data, which is severely scarce among new users. For example, in recommendations for new users on news feed platforms, the lack of sufficient behavioral logs and the resulting empty or extremely sparse behavioral features prevent the recommendation system from effectively capturing the potential interests of new users. Consequently, the recommendation results have a low degree of matching with the user's actual preferences, impacting the overall recommendation effectiveness.

[0022] In view of this, this disclosure proposes a completely new approach. To facilitate understanding of the embodiments of this disclosure, a brief introduction to the system to which this disclosure applies will first be given. For example... Figure 1 As shown, the system may include user equipment and a resource recommendation device located on the server side.

[0023] User devices can include, but are not limited to, smart mobile terminals, smart home devices, wearable devices, and PCs (Personal Computers). Smart mobile devices can include mobile phones, tablets, laptops, PDAs (Personal Digital Assistants), and connected cars. Smart home devices can include smart TVs, smart refrigerators, and so on. Wearable devices can include smartwatches, smart glasses, virtual reality devices, augmented reality devices, and mixed reality devices (i.e., devices that support both virtual and augmented reality).

[0024] The resource recommendation device can recommend resources to target users using user equipment using the methods provided in the embodiments of this disclosure.

[0025] The resource recommendation device can be configured as a standalone server, within a server cluster, or on a cloud server. A cloud server, also known as a cloud computing server or cloud host, is a hosting product within the cloud computing service system, designed to address the management difficulties and weak service scalability inherent in traditional physical hosts and Virtual Private Servers (VPS) services. In addition to... Figure 1 In addition to the architecture shown, the resource recommendation device can also be set up on a computer terminal with strong computing power.

[0026] It should be understood that Figure 1 The number of user devices and resource recommendation devices shown is merely illustrative. Any number of user devices and resource recommendation devices can be included depending on implementation needs.

[0027] Figure 2 A flowchart of a resource recommendation method provided in this disclosure embodiment, the method can be... Figure 1 The resource recommendation device in the system shown is executed. For example... Figure 2 As shown, the method may include the following steps: Step 201: Determine the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs.

[0028] Step 203: Obtain the preference intensity information of the target category under the target channel. The target channel is the channel from which the target user obtains the target application.

[0029] Step 205: Adjust the ranking score of the resource to be recommended using preference intensity information.

[0030] Step 207: Based on the adjusted ranking score, determine the resources to recommend to the target user.

[0031] As can be seen from the above technical solutions, by utilizing prior data such as the channel information from which the target user obtains the target application, the preference intensity information of the target category under the target channel is obtained, and the score of the resource to be recommended is adjusted accordingly during the ranking stage. This enables the recommendation results to match the potential interests and preferences of the target user related to the channel, solving the problem of mismatch between recommendation results and user interests caused by the lack of user behavior characteristics during the cold start stage, and improving the recommendation effect.

[0032] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments. It should be noted that the terms "first" and "second" used in this disclosure do not have limitations in terms of size, order, or quantity, but are only used to distinguish them by name. For example, "first screening condition" and "second screening condition" are only used to distinguish the two screening conditions by name.

[0033] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0034] First, the above step 201, namely "determining the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs", will be described in detail with reference to the embodiments.

[0035] In this embodiment of the disclosure, the resources to be recommended refer to various resources that the resource distribution platform plans to distribute to users, and can also be candidate resources. Examples include text and images in an information stream, videos in a short video stream, and products in an e-commerce platform.

[0036] For each resource to be recommended, the application from which it originates (i.e., the target application) is identified. This is usually the same application originating from the user client that initiated the resource recommendation request. More importantly, it is necessary to determine the content category (i.e., the target category) to which the resource to be recommended belongs. The category involved in this disclosure can be a first-level category, a second-level category, or a more granular category. For example, a short video titled "Must-See for Beginners: Top Ten Classic Single-Player Games" might have a target category of "Games" or "Game Reviews".

[0037] In one implementation, the metadata corresponding to each resource to be recommended contains a pre-labeled category ID. The server can obtain the target category to which the resource to be recommended belongs by querying the metadata.

[0038] The following describes step 203, namely "obtaining the preference intensity information of the target category under the target channel, where the target channel is the channel from which the target user obtains the target application," in detail with reference to the embodiments.

[0039] After determining the target category of the resources to be recommended, the system further obtains a key prior information: the preference strength of the target category under the target channel.

[0040] The target channel refers to the source through which the target user acquired the target application. When a user installs an application, the attribution system records the source channel. This could be a pre-installed feature on a specific phone brand, such as "pre-installed on phone brand A"; it could be downloaded and installed via a referral from another application, such as "application 1 - referral"; or it could be downloaded and installed from an app store, such as "app market B". This channel information is associated with the user's identifier (User_ID) and stored in the user profile system.

[0041] The preference intensity information obtained in this step is a quantitative indicator used to characterize whether the user group that acquires the application through this target channel has a stronger or weaker preference for resources in a specific category relative to the overall average level.

[0042] In the online service process of resource recommendation, when it is necessary to rank and score a resource to be recommended, the system first identifies the target user who initiated the request and obtains their attribution channel, i.e., the target channel. Next, based on the target category of the resource to be recommended determined in the previous steps, it obtains a pre-calculated preference intensity information specific to that "target channel-target category" combination. For example, if the target user comes from the "pre-installed on mobile phone brand A" channel, and the resource to be recommended belongs to the "anime" category, then the preference intensity information for the "anime" category under the "pre-installed on mobile phone brand A" channel is obtained.

[0043] One feasible approach is to obtain the preference intensity information of the target category under the target channel in real time in response to the resource recommendation request of the target user. However, to avoid the delay caused by online statistical calculations during the recommendation process, a mapping data can be pre-built and maintained. This mapping data can use the channel identifier and category identifier as a joint index to store the pre-determined preference intensity values ​​of each category under each channel.

[0044] In this step, based on the target category and target channel, we can query the preference intensity information of each pre-determined category under each channel, i.e., the above-mentioned mapping data, to obtain the preference intensity information of the target category under the target channel; wherein, the categories in the mapping data include the target category, and the channels include the target channel.

[0045] For example, when it is necessary to obtain the preference intensity of a target category (e.g., "games") under a target channel (e.g., "Huawei-pre-installed-CPI settlement"), you only need to use the identifier of the target channel and the identifier of the target category as query keys to directly retrieve the mapping data and return the corresponding preference intensity information.

[0046] By pre-constructing "channel-category" preference intensity information and directly querying and retrieving it during recommendation, the calculation operation is separated from the online recommendation process and converted into an offline preprocessing task. This solves the high latency problem caused by online real-time calculation and effectively reduces the computational load and response latency of the online recommendation service.

[0047] As one possible approach, the preference intensity information of a target category under a target channel is obtained based on statistical data of the target category's historical user behavior under the target channel. The statistical data may include impressions, clicks, and interactions.

[0048] Impressions refer to the number of times a resource of a specific category is shown to users on a specific channel; it represents the basic scale of resource content exposure. Clicks quantify the conversion behavior of users from passive browsing to active exploration, reflecting the attractiveness of the resource. Interactions further measure users' willingness to engage deeply, including but not limited to likes, favorites, comments, forwards, and shares; these behaviors usually reflect users' genuine interest and emotional inclination towards the resource more than clicks.

[0049] By using historical user behavior statistics to generate preference intensity information, the consumption behavior and interest intensity of user groups on specific categories under different channels are quantified, avoiding the bias caused by relying on a single behavioral indicator or subjective judgment. This allows preference intensity information to more accurately and stably reflect the true group preferences.

[0050] In addition, the aforementioned preference intensity information can also be obtained without relying on historical behavioral statistics, but through other means, such as obtaining popular category rankings from various channels and obtaining preference intensity information for the target category under the target channel based on popularity information.

[0051] One feasible approach is to use a target audience index to quantify the strength of user preference for a specific category on a specific channel. Specifically, the target audience index for a target category on a target channel is the ratio of the target category's interaction rate on the target channel to the overall interaction rate of all categories on the target channel.

[0052] Interaction rate is the ratio of the weighted sum of clicks and interactions to impressions. For example, target category. In target channels Interaction rate for: in, For target category In target channels Click count below For target category In target channels The number of interactions below For target category In target channels The number of impressions. The weighting coefficients are preset and can be empirical or experimental values.

[0053] All categories in the target channel Overall interaction rate for: Target Category In target channels The target group index below for: When the target group index is greater than 1, it indicates that the interaction performance of the target category is better than the channel average, and users have a stronger preference for it; less than 1 indicates a weaker preference.

[0054] Target group metrics, expressed as ratios, can eliminate the interference caused by differences in the scale of different channels, thus more objectively reflecting the relative popularity of a category within each channel. By weighting and integrating click behavior with interaction behavior, it can more comprehensively reflect users' willingness to engage than relying solely on click-through rate, especially when interaction volume (such as comments and shares) is more indicative of genuine preferences than clicks.

[0055] Besides the methods mentioned above, other approaches can be used to quantify the preference intensity of a target category on a target channel. For example, normalized interaction rates can be used to quantify the preference intensity of a target category on a target channel, or the ratio of the conversion rate of a target category on a target channel to the overall conversion rate of all categories on a target channel can be used to quantify the preference intensity of a target category on a target channel, and so on.

[0056] Furthermore, after determining the preference intensity information for each category under each channel, i.e., the "channel-category" preference intensity information, a preliminary filtering operation can be performed. Specifically, the filtering operation can be based on the following filtering conditions or combinations thereof: Retain the preference intensity information corresponding to channels that meet the first screening criteria and / or categories that meet the second screening criteria.

[0057] The first screening criterion is directly related to the scale of new users brought by the channel, that is, the first screening criterion is related to the number of new users or the proportion of new users acquired through the channel for the target application. For example, it can be set as follows: the number of unique visitors (UV) to the target application brought by the channel within a preset period (such as the past 14 days) is not less than a certain threshold, or the proportion of new users ranks among the top several, etc.

[0058] The second screening criterion focuses on the preference intensity of the category itself, meaning that the second screening criterion is related to the preference intensity information of each category under each channel. For example, it can be set that the Target Group Index (TGI) of a certain category under the channel is greater than a preset index threshold, or that the categories are sorted in descending order of TGI and the top few categories are selected. The categories obtained in this way are the strongly preferred categories under the channel.

[0059] By performing this filtering process, only the preference strength information of "channel-category" that has both a sufficient user base to ensure statistical significance and a clear preference tendency is retained. For example, suppose we obtain initial "channel-category" preference strength information for hundreds of channels and dozens of primary categories. Then, the first filtering condition is sorted by the percentage of new user UV, retaining only the top 10 mainstream channels and filtering out the preference strength information of all categories under other channels. Next, according to the second filtering condition, the top three categories with the highest TGI values ​​under each retained channel are retained, filtering out the preference strength information of other categories. The final result is the preference strength information of "channel-strong preference categories".

[0060] This approach effectively avoids the problem of data sparsity: the TGI value calculated for a channel with only a small number of new users may fluctuate wildly due to random factors, and using it as the basis for adjusting the ranking of subsequent recommended resources would introduce noise and lead to misjudgments. By filtering and retaining only statistically significant channels, the reliability of the recommendation results is improved. Furthermore, the selection of strongly preferred categories ensures that subsequent ranking adjustments for recommended resources focus on core categories that best reflect the characteristics of the channel, avoiding unnecessary ranking adjustments to non-critical categories. This improves recommendation relevance while maintaining the diversity of the recommendation results.

[0061] As can be seen from the above process, the process of pre-constructing the preference intensity information of "channel-category" can be as follows: Figure 3a As shown, data such as impressions, clicks, and interactions for each category across various channels are first collected to calculate the TGI for each category on each channel. After calculation, the preference intensity information for "New User UV Percentage Channels - Strong Preference Categories" is selected according to the first and second filtering criteria mentioned above. The preference intensity information for "New User UV Percentage Channels - Strong Preference Categories" can be saved as a mapping table, and the mapping table is updated periodically to maintain the timeliness of channel preferences.

[0062] The following describes step 205, namely "adjusting the ranking score of the resource to be recommended using preference intensity information," in detail with reference to an embodiment.

[0063] After obtaining the preference intensity information of the target category under the target channel, the preference intensity information is used as an adjustment factor to adjust the basic ranking score of the resource to be recommended, based on the basic ranking score. The original basic ranking score can be generated by a conventional ranking model, which may integrate multiple features such as content quality, user behavior feature matching degree, and timeliness. The method of obtaining the basic ranking score in this embodiment is not particularly limited.

[0064] The core logic of adjusting using preference strength information lies in correspondingly strengthening or weakening the ranking score of the resource to be recommended based on the degree of preference indicated by the value. For example, when preference strength information indicates that users of the target channel have a strong preference for the target category, the ranking score of the resource to be recommended is increased, allowing it to occupy a higher position in the final ranking. Conversely, if preference strength information indicates that users of the target channel have a weak preference for the target category, its ranking score is decreased. This adjustment method allows the recommendation results to reflect the prior group interests of users from the channel source, thus effectively influencing the ranking results when historical behavioral data of new users is lacking.

[0065] In this step, the ranking score of the resource to be recommended can be adjusted based on the adjustment factor. The value of the adjustment factor is positively correlated with the preference intensity information of the target category under the target channel, so that the ranking score of the resource to be recommended is increased when the preference intensity information is greater than the first intensity threshold, and the ranking score of the resource to be recommended is decreased when the preference intensity information is less than the first intensity threshold.

[0066] In other words, a ranking score adjustment mechanism is proposed that is positively correlated with preference intensity information and uses a first intensity threshold as the dividing point. This adjustment mechanism achieves differentiated enhancement or weakening of the basic ranking score through an adjustment factor that dynamically changes with preference intensity.

[0067] One feasible approach is to multiply the adjustment factor by the ranking score of the resource to be recommended, thus obtaining the adjusted ranking score. The adjustment factor is derived from the exponential operation of the fusion parameters, where the exponent is the difference between the preference intensity information of the target category under the target channel and a first intensity threshold. For example, the adjusted ranking score... The following formula can be used: in, The ranking score (i.e., the basic ranking score) is the score for the resource to be recommended. For the fusion parameters, experimental or empirical values ​​greater than 1 are used. Among these, higher values ​​are preferred. A value close to 1 makes resource recommendations rely more heavily on channel priors (i.e., the strength of preference for the target category under the target channel), suitable for stages with a strong need for new user retention but tolerating a decrease in diversity; while a value close to 1... The value makes resource recommendations smoother and more conservative, suitable for scenarios with a small number of new users. This refers to the target audience index for the target category within the target channel.

[0068] For example, suppose the fusion parameters The value is 2. The current user comes from the "Pre-installed on mobile phone brand A" channel. The category of the resource to be recommended is "Anime," and the TGI value for this category under this channel is 2.5. Therefore, the adjustment factor is... The adjusted ranking score is approximately 2.828 times the base ranking score. Conversely, if the TGI value for this category is 0.6, the adjustment factor is... The adjusted ranking score is compressed to 0.758 times the base ranking score. When the TGI is exactly 1, the adjustment factor is 1, and the score remains unchanged. This exponential function design makes the ranking adjustment magnitude change non-linearly with the degree to which the TGI deviates from 1: when the preference intensity is much higher than average, the adjustment factor increases sharply, providing a strong gain; when the preference intensity is much lower than average, the adjustment factor decreases rapidly, imposing a significant deweighting.

[0069] As can be seen, the exponential function is mathematically smooth and monotonically increasing (when the base is greater than 1), ensuring that the adjustment factor changes continuously with the intensity of preference, avoiding possible jumps, and thus guaranteeing a smooth transition in the adjusted ranking results. Furthermore, because the exponential function amplifies differences above a threshold, it can more significantly distinguish differences between different levels of preference, making the recommendation results more sensitive to preference signals. Simultaneously, it rapidly decays differences below the threshold, effectively suppressing the display opportunities for non-preference categories. In addition, the introduction of the fusion parameter provides an adjustable hyperparameter, allowing for flexible adjustment of the fusion parameter value based on online experimental results to control the overall strength of the adjustment strategy.

[0070] Besides exponential functions, other types of continuous functions can also be used to construct adjustment factors, such as linear functions, like adjustment factor = 1 + k(TGI-1).

[0071] In this embodiment, the aforementioned ranking score adjustment can be performed on all resources to be recommended. However, this approach may be too aggressive in downgrading resources in weakly preferred categories, leading to recommendations that deviate from actual user needs. Therefore, as a better approach, a condition for adjusting the ranking score can be introduced. The adjustment logic will only be triggered if the categories are pre-determined to be content that users have a significant preference for under the target channel. That is, if the target category belongs to a strongly preferred category under the target channel, the ranking score of the resource to be recommended will be adjusted using the preference intensity information. As described in the previous embodiments, a strongly preferred category can be a category under the target channel whose corresponding preference intensity information ranks in the top N and / or is greater than a preset intensity threshold, where N is a positive integer.

[0072] This mechanism focuses on the scope of ranking score adjustments, ensuring that adjustments are concentrated on resources most likely to improve user satisfaction, thus avoiding the introduction of noise or reduced recommendation quality due to inappropriate adjustments to weakly favored categories. Furthermore, in recommendation systems, the resource pool containing the resources to be recommended typically contains a large number of resources; performing adjustments on each resource would accumulate a considerable computational burden. This embodiment adjusts ranking scores only for strongly favored categories, avoiding most invalid computations and improving the execution efficiency and throughput of the recommendation system.

[0073] In this embodiment, the effectiveness strategy for adjusting ranking scores can be further controlled. This mainly includes controlling the target population and the timing of the effect.

[0074] The control strategy for the target audience can be designed to apply only to new users. For example, if the target users are new users of the target application, the ranking score of the resource to be recommended can be adjusted using preference intensity information.

[0075] The definition of a new user is at least related to the sparsity of user behavior in the target application. Users whose behavior sparsity meets preset sparsity requirements can be considered new users. Specifically, a new user can be a user who meets at least one of the following conditions: Condition 1: The resource recommendation service of the target application is used within a preset time period.

[0076] This condition actually defines new users (i.e., cold start scenarios) from a time dimension. For example, in terms of time, a preset duration of 14 days can be set, meaning that the above-mentioned ranking score adjustment is only applied to users who have used the recommendation service for no more than 14 days since their first use (such as "feed14-day new users").

[0077] Condition 2: The number of preset type interactive behaviors in the target application is within a preset number.

[0078] This condition actually defines new users (i.e., cold start scenarios) from a behavioral perspective. For example, in terms of behavior, a preset number of clicks can be set to 5, and adjustments can only be made for users who have clicked less than 5 times in the target application.

[0079] These two conditions can be used individually or in combination. For example, a stricter screening criterion can be set: only users who simultaneously meet the criteria of "using the recommendation service for less than 14 days" and "having fewer than 5 historical clicks" are considered cold-start users requiring ranking score adjustments based on preference strength information. Conversely, if a user's usage time exceeds 14 days, or their interaction count reaches 5 or more, it is considered that the user has accumulated sufficient personalized behavioral data to generate ranking results using a conventional recommendation model, and therefore automatically exits the mechanism for ranking score adjustments based on preference strength information.

[0080] It can be seen that by controlling the target user group, the ranking score adjustment mechanism based on preference intensity information is accurately deployed and strictly limited to the cold start phase. This avoids the incorrect application of this adjustment mechanism to mature users who already have stable personal interest models, preventing unnecessary disturbances to the recommendation results of mature users, thereby ensuring the stability of the recommendation system and the user experience. Secondly, as users accumulate their own behavior data, when the amount of behavioral data reaches a certain level, the exit mechanism is automatically triggered, returning the ranking control to the regular recommendation model based on their personal interests. This achieves a natural exit of the cold start strategy, which conforms to the objective laws of user interest evolution.

[0081] The timing of the recommendations can be controlled to apply only to the first few resource recommendations in the current session. For example, if the number of resource recommendations for the target user in the current session is less than or equal to a preset threshold, the ranking score of the recommended resource can be adjusted using preference intensity information.

[0082] The current session refers to a continuous interaction cycle from when a target user initiates a recommendation request by opening the target application until the application is closed or reactivated after running in the background for a long time. For example, in a news feed service, a user's action of swiping up and down to load new resources constitutes a recommendation request. In a typical implementation scenario, the preset threshold for the number of requests can be set to 3. This means that in a session, the mechanism for adjusting the ranking score based on preference intensity information will only be applied during the user's first 3 refresh requests; from the 4th refresh onwards, the regular ranking model will be used again.

[0083] It can be seen that in the early stages of a session, when the target user has not yet generated any behavioral feedback within the session, channel preference (i.e., the preference strength information of the target category under the target channel) is used for strong guidance, optimizing the user's initial consumption experience. In the later stages of the session, since the mechanism of adjusting the ranking score based on preference strength information automatically becomes invalid, the target user has the opportunity to be exposed to resources from more different categories, thereby expanding their interest boundaries and avoiding the "information cocoon" effect caused by repeatedly recommending the same category of resources for a long time.

[0084] Besides the two activation strategies mentioned above, other activation strategies can also be used, which will not be listed here.

[0085] The following describes step 207, namely "determining resources to recommend to the target user based on the adjusted ranking score," in detail with reference to an embodiment.

[0086] In the aforementioned embodiments, the ranking score of each resource to be recommended has been adjusted based on the preference intensity information of its target category under the target channel. Resources with an adjusted ranking score greater than or equal to a preset score threshold can be selected as resources recommended to the target user. Alternatively, the resources can be ranked based on the adjusted ranking score, and the top-ranked resources can be selected as resources recommended to the target user. This ensures that high-quality resources that best match the source channel preferences of the target user are presented to the user first, thereby improving the cold start experience for new users.

[0087] Furthermore, it can monitor the cumulative behavioral data of target users in the target application; if the preset exit conditions are met, it will stop using preference intensity information to adjust the ranking score of the recommended resources and switch to the regular recommendation process.

[0088] The aforementioned preset exit conditions may include at least one of the following: the cumulative time the target user uses the resource recommendation service of the target application exceeds a preset time threshold; or the cumulative number of preset type of interactive behaviors performed by the target user in the target application exceeds a preset number threshold.

[0089] The aforementioned dynamic exit mechanism ensures that the channel-priority-based recommendation strategy only plays a dominant role in the initial cold start phase when user behavior data is sparse. Once users' behavior provides sufficient data to construct user characteristics, the system smoothly transitions to the regular recommendation process, thus avoiding the problem of content homogenization in recommendation results caused by long-term reliance on channel priority.

[0090] The methods provided in this disclosure can be widely applied to, but are not limited to, the following typical scenarios: Multi-channel customer acquisition and recommendation scenarios. Platforms such as news feeds, short video feeds, and e-commerce platforms typically have highly dispersed user sources, encompassing various channels including pre-installation by manufacturers, app store promotions, social media placements, and content platform traffic. New users from different channels often have significantly different initial interests and preferences. By associating each new user with the channels they used to acquire the application and adjusting the ranking scores of recommended resources based on offline statistical preference data from those channels, the application's server can configure differentiated recommendations for new users from different channels without relying on any user history. For example, users from game promotion channels prioritize seeing game and anime resources; users from knowledge community channels prioritize seeing science and humanities resources. This capability effectively bridges the gap between customer acquisition channels and content consumption in multi-channel customer acquisition scenarios, improving the overall recommendation effect for cross-channel users.

[0091] The core application scenario of this disclosure is the cold start recommendation phase for new users. New users lack historical behavioral data, and recommendation services have long faced the dilemma of "how to provide personalized content with little or no user behavioral data." The method provided in this disclosure successfully constructs a personalized recommendation approach that does not rely on accumulated user behavior, significantly improving the content consumption experience for new users during the cold start phase.

[0092] In recommendation services that are sensitive to the first-screen experience, a user's initial experience directly determines whether they are willing to continue using the product. The activation control mechanism proposed in this disclosure (such as limiting the activation to low-click new users and limiting the activation time to the first few refreshes of the session) ensures that channel preferences take effect at the moment when users most need guidance and when the risk of system intervention is lowest. By prioritizing the display of highly matched resources enhanced by channel preferences in the first few interactions of each time a user opens the application, resources that match the user's preferences can be accurately recommended before the user invests any learning costs.

[0093] The following is combined Figure 3b The process shown illustrates an implementation example of a new user cold start phase applied to a media streaming application.

[0094] In this embodiment, the resource recommendation device is deployed on the server side of the media streaming application. When user a opens the application (e.g., in client form) and enters the feed stream, a feed stream request is initiated. The resource recommendation device first determines whether user a is a new user. In this embodiment, if user a has been using the application for less than 14 days and has fewer than 5 historical clicks, then user a is identified as a new user.

[0095] If the user is not a new user, the standard recommendation process will be followed, for example, using the ranking score of the media file to be recommended (i.e., the base ranking score) to determine the media file to be recommended to user a.

[0096] If the user is new, determine if the current recommendation is one of the first three recommendations in the current session. If not, proceed with the regular recommendation process. If so, retrieve the channel information for user 'a' to acquire the application, i.e., the target channel information. This information is recorded and stored when the user acquires the application, such as "Mobile phone brand A - pre-installed - CPI settlement".

[0097] The system determines whether the target channel is among the top 10 channels in terms of new user UV. If not, it determines the media files to recommend to user a based on the base ranking score. It queries the pre-calculated and stored "channel-strong preference category" preference strength information (e.g., TGI) to determine if the category of the media to be recommended belongs to the target channel's strong preference category. If it does, the ranking score (i.e., the base ranking score) of the media to be recommended is adjusted using the preference strength information; otherwise, the ranking score remains unchanged. Finally, the resource recommendation device sorts all the media to be recommended from highest to lowest according to the adjusted ranking score (or the base ranking score if not adjusted), selects the top few items with the highest scores as the recommendation results, and sends them to the user's terminal for display.

[0098] As can be seen from the above embodiments, the resource recommendation device achieves personalized recommendation intervention for new users during the cold start phase based on prior information from attribution channels.

[0099] Figure 4 A schematic block diagram of the resource recommendation device provided in the embodiments of this disclosure, such as Figure 4 As shown, the device may include: a determining unit 401, an acquiring unit 402, an adjusting unit 403, and a recommending unit 404, and also includes a filtering unit 405. The main functions of each component are as follows: The determining unit 401 is configured to determine the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs.

[0100] The acquisition unit 402 is configured to acquire the preference intensity information of the target category under the target channel, where the target channel is the channel from which the target user acquires the target application.

[0101] Adjustment unit 403 is configured to adjust the ranking score of the resource to be recommended using preference intensity information.

[0102] Recommendation unit 404 is configured to determine resources to recommend to the target user based on the adjusted ranking score.

[0103] As one possible implementation method, the acquisition unit 402 can be specifically configured to: based on the target category and the target channel, query the preference intensity information of each pre-determined category under each channel, and obtain the preference intensity information of the target category under the target channel; wherein, each category includes the target category, and each channel includes the target channel.

[0104] In some embodiments, the preference intensity information of the target category under the target channel is obtained based on statistical data of the historical user behavior of the target category under the target channel. The statistical data includes impressions, clicks, and interactions.

[0105] As one possible approach, preference intensity information is the target group index; the target group index of the target category under the target channel is the ratio of the interaction rate of the target category under the target channel to the overall interaction rate of all categories under the target channel; where the interaction rate is the ratio of the weighted sum of clicks and interactions to impressions.

[0106] Furthermore, the filtering unit 405 can be configured to: before the acquisition unit 402 acquires the preference intensity information of the target category under the target channel, perform a filtering operation on the preference intensity information of each predetermined category under each channel, so as to retain the preference intensity information corresponding to the channel that meets the first filtering condition and / or the category that meets the second filtering condition; wherein the first filtering condition is related to the number of new users or the proportion of new users of the target application acquired through the channel; and the second filtering condition is related to the preference intensity information of the category under each channel.

[0107] As one possible implementation method, the adjustment unit 403 can be specifically configured to: adjust the ranking score of the resource to be recommended based on the adjustment factor, wherein the value of the adjustment factor is positively correlated with the preference intensity information of the target category under the target channel, so that when the preference intensity information is greater than the first intensity threshold, the ranking score of the resource to be recommended is increased, and when the preference intensity information is less than the first intensity threshold, the ranking score of the resource to be recommended is decreased.

[0108] Specifically, when adjusting the ranking score of the resource to be recommended based on the adjustment factor, the adjustment unit 403 can be configured to: obtain the adjusted ranking score by multiplying the adjustment factor by the ranking score of the resource to be recommended; wherein the adjustment factor is obtained by the exponential operation of the fusion parameter, the exponent used in the exponential operation is the difference between the preference intensity information of the target category under the target channel and the first intensity threshold, and the fusion parameter is a value greater than 1.

[0109] As another possible implementation, the adjustment unit 403 can be specifically configured as follows: if the target category belongs to the strong preference category corresponding to the target channel, the ranking score of the resource to be recommended is adjusted using the preference intensity information; wherein, the strong preference category is the category under the target channel whose corresponding preference intensity information ranks in the top N and / or is greater than the preset intensity threshold, and N is a positive integer.

[0110] As another possible implementation, the adjustment unit 403 can be specifically configured to: if the target user is a new user of the target application, adjust the ranking score of the resource to be recommended using preference intensity information; wherein, the new user is a user who meets at least one of the following conditions: uses the resource recommendation service of the target application within a preset time period; and engages in a preset type of interaction behavior within the target application within a preset number of times.

[0111] As another possible implementation, the adjustment unit 403 can be specifically configured to: if the number of resource recommendations for the target user in the current session is less than or equal to a preset number threshold, then adjust the ranking score of the resource to be recommended using preference intensity information.

[0112] The collection, storage, use, processing, transmission, provision, and disclosure of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0113] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0114] Figure 5 A schematic block diagram of an example electronic device 500 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.

[0115] like Figure 5As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 502 or a computer program loaded from storage unit 508 into random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. Input / output (I / O) interface 505 is also connected to bus 504.

[0116] Multiple components in device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0117] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 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 501 performs the various methods and processes described above, such as the resource recommendation method. For example, in some embodiments, the resource recommendation method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of the resource recommendation method described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the resource recommendation method by any other suitable means (e.g., by means of firmware).

[0118] 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.

[0119] 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.

[0120] 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.

[0121] 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).

[0122] 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.

[0123] 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, servers in distributed systems, or servers incorporating blockchain technology.

[0124] 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.

[0125] 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 resource recommendation method, characterized in that, include: Determine the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs; Obtain the preference intensity information of the target category under the target channel, where the target channel is the channel from which the target user obtains the target application; The ranking score of the resource to be recommended is adjusted using the preference intensity information; Based on the adjusted ranking score, resources recommended to the target user are determined.

2. The method according to claim 1, wherein, The step of obtaining the preference intensity information of the target category under the target channel includes: Based on the target category and the target channel, the preference intensity information of each predetermined category under each channel is queried to obtain the preference intensity information of the target category under the target channel; wherein, each category includes the target category, and each channel includes the target channel.

3. The method according to claim 1, wherein, The preference intensity information of the target category under the target channel is obtained based on the statistical data of the historical user behavior of the target category under the target channel. The statistical data includes impressions, clicks and interactions.

4. The method according to claim 3, characterized in that, The preference intensity information is a target group index; The target group index of the target category under the target channel is the ratio of the interaction rate of the target category under the target channel to the overall interaction rate of all categories under the target channel; The interaction rate is the ratio of the weighted sum of clicks and interactions to impressions.

5. The method according to claim 2, further comprising, before obtaining the preference intensity information of the target category under the target channel: Perform a filtering operation on the preference intensity information of each predetermined category under each channel to retain the preference intensity information corresponding to the channels that meet the first filtering condition and / or the categories that meet the second filtering condition; The first filtering condition is related to the number of new users or the percentage of new users of the target application obtained through the channel; the second filtering condition is related to the preference intensity information of each category under each channel.

6. The method according to any one of claims 1 to 5, wherein, The step of adjusting the ranking score of the resource to be recommended using the preference intensity information includes: The ranking score of the resource to be recommended is adjusted based on an adjustment factor, wherein the value of the adjustment factor is positively correlated with the preference intensity information of the target category under the target channel, so that when the preference intensity information is greater than a first intensity threshold, the ranking score of the resource to be recommended is increased, and when the preference intensity information is less than the first intensity threshold, the ranking score of the resource to be recommended is decreased.

7. The method according to claim 6, wherein, The adjustment of the ranking score of the resource to be recommended based on the adjustment factor includes: The adjusted ranking score is obtained by multiplying the adjustment factor by the ranking score of the resource to be recommended; wherein the adjustment factor is obtained by exponential calculation of the fusion parameter, the exponent used in the exponential calculation is the difference between the preference intensity information of the target category under the target channel and the first intensity threshold, and the fusion parameter is a value greater than 1.

8. The method according to any one of claims 1 to 5, wherein, The step of adjusting the ranking score of the resource to be recommended using the preference intensity information includes: If the target category belongs to the strong preference category corresponding to the target channel, the ranking score of the resource to be recommended is adjusted using the preference intensity information; wherein, the strong preference category is the category under the target channel whose corresponding preference intensity information ranks in the top N and / or is greater than a preset intensity threshold, and N is a positive integer.

9. The method according to any one of claims 1 to 5, wherein, The step of adjusting the ranking score of the resource to be recommended using the preference intensity information includes: If the target user is a new user of the target application, the ranking score of the resource to be recommended is adjusted using the preference intensity information; wherein, the new user is a user whose behavior sparsity in the target application meets the preset sparsity requirement.

10. The method according to any one of claims 1 to 5, wherein, The step of adjusting the ranking score of the resource to be recommended using the preference intensity information includes: If the number of resource recommendations for the target user in the current session is less than or equal to a preset threshold, the ranking score of the resource to be recommended is adjusted using the preference intensity information.

11. A resource recommendation device, characterized in that, include: The determining unit is configured to determine the target application to which the resource to be recommended belongs and the target category to which the resource to be recommended belongs; The acquisition unit is configured to acquire preference intensity information of the target category under the target channel, wherein the target channel is the channel from which the target user acquires the target application; The adjustment unit is configured to adjust the ranking score of the resource to be recommended using the preference intensity information; The recommendation unit is configured to determine the resources to recommend to the target user based on the adjusted ranking score.

12. 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 executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method of any one of claims 1-10.

13. 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 1-10.

14. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-10.

Images