Map page display method and device and computing equipment

Abstract

The embodiment of the invention provides a map page display method and device and computing equipment, and the method comprises the steps: responding to a map page display event, and obtaining a target triggering scene corresponding to the map page display event; obtaining target historical behavior data in the target triggering scene; according to the target triggering scene and / or the target historical behavior data, determining a recommendation object in the target triggering scene from candidate distribution objects corresponding to the map page; and distributing the recommendation object to the client to enable the client to display a map page, and displaying the recommendation object in the target triggering scene in the map page. The recommendation objects which the user may be interested in under the current triggering scene are recommended to the user, the requirement of the user in the current triggering scene is better met, personalized recommendation of different triggering scenes is achieved, and user operation is facilitated.

Description

Technical Field

[0001] The embodiments of this specification relate to the field of computer technology, and more particularly to a map page display method, apparatus, and computing device. Background Art

[0002] With the rapid development of computer and Internet technologies, a variety of content services can be provided through content application platforms, such as map exploration services. The content application platform can provide a map page. When users enter the map page, the recommended nearby points (markers), such as businesses, services, and attractions, can be displayed on the map based on the user's current location. This meets the user's map exploration needs for travel, LBS (Location-Based Services), social networking, and other purposes, and realizes the connection from online to offline.

[0003] In the prior art, recommended points on a map page are sorted and displayed according to default rules. The recommended points displayed when each user enters the map page are consistent. The conversion rate of the recommended points displayed on the map page is low, resulting in low user stickiness of the map page. Summary of the Invention

[0004] In view of this, embodiments of this specification provide a map page display method. One or more embodiments of this specification also relate to another map page display method, a map page display device, another map page display device, a computing device, a computer-readable storage medium, and a computer program product to address technical deficiencies in the prior art.

[0005] According to a first aspect of an embodiment of this specification, a map page display method is provided, which is applied to a server and includes:

[0006] In response to a map page display event, obtaining a target triggering scene corresponding to the map page display event;

[0007] Obtaining historical target behavior data under the target triggering scenario;

[0008] Determining a recommended object under the target triggering scenario from candidate distribution objects corresponding to the map page according to the target triggering scenario and / or the target historical behavior data;

[0009] The recommended objects are distributed to a client, so that the client displays the map page and displays the recommended objects in the target triggering scenario on the map page.

[0010] According to a second aspect of an embodiment of this specification, a map page display method is provided, which is applied to a client and includes:

[0011] In response to a map page display event, obtaining a recommended object corresponding to a target triggering scenario of the map page display event, wherein the recommended object is determined from candidate distribution objects corresponding to the map page based on the target triggering scenario and / or target historical behavior data under the target triggering scenario;

[0012] The map page is displayed, and the recommended objects are displayed on the map page.

[0013] According to a third aspect of the embodiments of this specification, a map page display device is provided, which is applied to a server, including:

[0014] A first acquisition module is configured to, in response to a map page display event, acquire a target triggering scene corresponding to the map page display event;

[0015] A second acquisition module is configured to acquire target historical behavior data in the target triggering scenario;

[0016] a determination module configured to determine, based on the target triggering scenario and / or the target historical behavior data, a recommended object under the target triggering scenario from among the candidate distribution objects corresponding to the map page;

[0017] The distribution module is configured to distribute the recommended objects to the client, so that the client displays the map page and displays the recommended objects in the target triggering scenario on the map page.

[0018] According to a fourth aspect of the embodiments of this specification, a map page display device is provided, which is applied to a client and includes:

[0019] a third acquisition module configured to, in response to a map page display event, acquire a recommended object corresponding to a target triggering scenario of the map page display event, wherein the recommended object is determined from candidate distribution objects corresponding to the map page based on the target triggering scenario and / or the target historical behavior data under the target triggering scenario;

[0020] The display module is configured to display the map page and display the recommended objects on the map page.

[0021] According to a fifth aspect of the embodiments of this specification, there is provided a computing device, including:

[0022] memory and processor;

[0023] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the above-mentioned map page display method are implemented.

[0024] According to a sixth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores computer-executable instructions, and when the instructions are executed by a processor, the steps of the above-mentioned map page display method are implemented.

[0025] According to a seventh aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above-mentioned map page display method when executed by a processor.

[0026] In an embodiment of the present specification, a map page display method is provided, which responds to a map page display event, obtains a target triggering scenario corresponding to the map page display event; obtains target historical behavior data under the target triggering scenario; determines a recommended object under the target triggering scenario from each candidate distribution object corresponding to the map page based on the target triggering scenario and / or the target historical behavior data; distributes the recommended object to a client, so that the client displays the map page, and displays the recommended object under the target triggering scenario on the map page.

[0027] One embodiment of the present specification implements that, in response to a map page display event, a target triggering scenario that triggers the map page display event can be obtained, and based on the target historical behavior data under the target triggering scenario, a recommended object under the target triggering scenario is determined from each candidate distribution object corresponding to the map page, so that the client displays the recommended object under the target triggering scenario on the map page. In this way, based on the triggering scenario of the map page display event, the recommended objects to be displayed on the map page can be screened. When a user enters the map page under different triggering scenarios, the recommended objects displayed are different. The recommended objects that the user may be interested in under the current triggering scenario can be recommended to the user, thereby achieving personalized recommendations for different triggering scenarios, better meeting the user's needs in the current triggering scenario, facilitating user operation, improving the user experience, and thereby improving the conversion rate of recommended objects on the map page, ensuring user stickiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 This is a flow chart of a map page display method provided by one embodiment of this specification;

[0029] Figure 2 This is an interface diagram of a content application platform provided by an embodiment of this specification;

[0030] Figure 3 This is a schematic diagram of a content details page for publishing content in a content application platform provided by an embodiment of this specification;

[0031] Figure 4 This is a schematic diagram of a process for determining a recommended object provided by an embodiment of this specification;

[0032] Figure 5 This is a schematic diagram of a trigger probability prediction process based on a target recommendation model provided by an embodiment of this specification;

[0033] Figure 6 is a flowchart of another map page display method provided by one embodiment of this specification;

[0034] Figure 7 is a schematic diagram of a first type of map page provided in one embodiment of this specification;

[0035] Figure 8 is a schematic diagram of a second type of map page provided in one embodiment of this specification;

[0036] Figure 9 This is a schematic diagram of a third type of map page provided in one embodiment of this specification;

[0037] Figure 10 This is a timing diagram of the processing process of a map page display method provided by one embodiment of this specification;

[0038] Figure 11 This is a structural diagram of a map page display device provided by one embodiment of this specification;

[0039] Figure 12 This is a schematic structural diagram of another map page display device provided by an embodiment of this specification;

[0040] Figure 13 This is a structural block diagram of a computing device provided by one embodiment of this specification. DETAILED DESCRIPTION

[0041] The following description sets forth many specific details to facilitate a thorough understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the scope of this specification. Therefore, this specification is not limited to the specific implementations disclosed below.

[0042] The terms used in one or more embodiments of this specification are for the purpose of describing specific embodiments only and are not intended to limit one or more embodiments of this specification. The singular forms "a," "an," and "the" used in one or more embodiments of this specification and the appended claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used in one or more embodiments of this specification refers to and includes any or all possible combinations of one or more associated listed items.

[0043] It should be understood that although the terms first, second, etc. may be used to describe various information in one or more embodiments of this specification, such information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of this specification, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0044] In addition, 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, stored data, displayed data, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0045] First, the terms involved in one or more embodiments of this specification are explained.

[0046] POI (Point of Interest): refers to predicting and recommending places or locations that users may be interested in based on their historical behavior, preferences, social relationships, and other information. This technology has a wide range of applications in fields such as tourism, catering, and entertainment. It can help users discover new places and improve user experience and satisfaction.

[0047] Marker: Usually refers to a marking point on a map, used to identify a specific location or place. Marker is a very common concept in geographic information systems (GIS), online map services (such as Google Maps, Amap), and location-based services (LBS).

[0048] LBS (Location-Based Services) social networking refers to an application that leverages a user's location information to enhance social networking services. It combines geolocation technology with social networking features, enabling people to interact, share, and discover new social opportunities based on their location. LBS social applications use the user's mobile device to obtain their current location and use this information to provide a range of personalized services.

[0049] UV: refers to "Unique Visitor", which is an important indicator used to count the number of different users who visit a website or application within a certain period of time. Each UV represents a unique user, and no matter how many times the user visits within the statistical period, it is only counted once.

[0050] Feed: Typically referred to as a stream of information or dynamic news, it is a core component of a social platform, showcasing the latest activities, published content, and other relevant information of users and those they follow. For LBS (location-based services) social apps, feeds not only include traditional content such as text, images, and videos, but can also incorporate location-related elements to provide a richer and more personalized user experience.

[0051] It should be noted that the map page of the content application platform is not only committed to meeting users' existing traditional map needs such as travel, LBS social networking, and realizing the connection from online to offline, but also hopes to create a new field that is different from the strong tool attribute maps in traditional industries. With real POIs and roads as the framework, rich content is mounted to show users the content they are interested in in the current time and space scenario, realizing the connection from content to users. The current map page does not have any recommendation and distribution capabilities. The content that users see when entering the map page is stereotyped. Markers such as POIs and group chats are sorted according to rules. The map page has nearly one million UVs per day, but only half of the users take any action. In order to improve user stickiness and click-through conversion rate on the map page, build a distribution field parallel to the feed, and explore new scenarios for content distribution, social networking, and games based on geographic location, the map page needs to strengthen the iterative construction of recommendation and distribution strategies.

[0052] In actual implementation, offline mining of a large amount of user information across three scenario dimensions, namely time, space, and status, as well as historical operation information on candidate POIs (such as clicks, favorites, etc.), is used to train the initial recommendation model. The target recommendation model, after training, can score each candidate POI based on the currently input information across three scenario dimensions, namely time, space, and status, select the recommended objects for that user at the current time, space, and status, and mark them on the map. Different users, or the same user, accessing the map page at different times, spaces, and states will display different recommended objects. Combining time, space, and status information, the system recommends the user the content that they are most interested in within their current location, time range, and state, achieving personalized recommendations and improving map page conversion rates and user stickiness.

[0053] In the embodiments of this specification, when a user enters the map page of a content application platform, he or she can see the markers recommended by the map page based on his or her own interest preferences, note browsing history, current time and space, and current actual travel / browsing status, including location markers, note markers, group chat markers, live broadcast markers, etc. That is to say, when a user enters the map page, the POI, group chat, note and other markers required by the user will be recommended based on the user's interest preferences, current time and space, status and other scenarios, and more importantly, the recommended markers will be in line with the triggering scenario for entering the map page.

[0054] In this specification, a map page display method is provided. This specification also involves a map page display device, a computing device, a computer-readable storage medium, and a computer program product, which are described in detail one by one in the following embodiments.

[0055] See also Figure 1 , Figure 1 A flowchart of a map page display method provided according to an embodiment of the present specification is shown, which is applied to a server and specifically includes the following steps.

[0056] Step 102: In response to a map page display event, obtain a target triggering scene corresponding to the map page display event.

[0057] Specifically, a map page display event refers to an event triggered by a user accessing a map page on a content application platform through a client. This event can be triggered by a click on a map page control on the content application platform, or by a location control included in published content on the content application platform. Of course, in actual implementation, other methods can also be used to access the map page and trigger the map page display event, and this embodiment of the specification does not limit this. Published content is a content service provided by the content application platform, such as notes provided by the content application platform.

[0058] For example, Figure 2 This is an interface diagram of a content application platform provided by an embodiment of this specification, such as Figure 2 As shown, the user clicks the "Nearby" control of the content application platform in the client to enter the recommendation page, which displays multiple "Nearby" notes. The user clicks the "Explore" control on the recommendation page, which triggers a map page display event and sends the map page display event to the server to display the map page in the client.

[0059] Another example, Figure 3 This is a schematic diagram of a content details page for publishing content in a content application platform provided by an embodiment of this specification, such as Figure 3As shown, the user can view the content details page of the published content on the content application platform of the client. The published content is a recommended note for a certain scenic spot. The content details page is the detailed recommendation information of the scenic spot, and the content details page includes a location control of the scenic spot. When the user clicks the location control, a map page display event is triggered, and the map page display event is sent to the server to display the map page in the client.

[0060] It should be noted that the target triggering scenario refers to the scenario in which the user triggers the map page display event, that is, the scenario in which the user enters the map page.

[0061] In actual implementation, the target trigger scenario may include a map page scenario, a trigger time and space, and the trigger time and space include a trigger time and / or a trigger space. Among them, the trigger time refers to the trigger time of the map page display event, that is, when the user enters the map page; the trigger space may refer to the trigger position, trigger status, etc. of the map page display event, that is, at what position and in what state the user enters the map page. The status may refer to the user's travel status or exploration status. The user's travel status may be normal, business trip, tourism, returning home, etc. The exploration status refers to what type of map page content the user wants to explore, such as the same city, different place, etc. The user's travel status can be obtained based on the current status configured by the user, and can be read from the user's attribute information or basic resources. The exploration type can be determined based on the type of triggering operation of the map page display event.

[0062] In an embodiment of the present specification, in response to a map page display event, the user of the specification wants to enter the map page. At this time, the server can obtain the target triggering scenario corresponding to the map page display event to determine the scenario in which the user enters the map page, thereby facilitating the subsequent targeted display of recommended objects on the map page based on the target triggering scenario.

[0063] Step 104: Obtain target historical behavior data in the target triggering scenario.

[0064] It should be noted that the target historical behavior data in the target triggering scenario can be obtained, based on which the historical preference in the target triggering scenario can be indicated, thereby facilitating the recommendation of the recommended object in the target triggering scenario.

[0065] In an optional implementation of this embodiment, the target triggering scene includes a map page scene and a triggering time and space;

[0066] Obtain historical target behavior data under target triggering scenarios, including:

[0067] Obtain relevant historical behavior data of the triggering user;

[0068] Target historical behavior data that meets the target trigger scenario is filtered out from the relevant historical behavior data, and the target historical behavior data includes historical distribution objects that the triggering user has interacted with and / or has not interacted with in the target trigger scenario.

[0069] It should be noted that a content application platform may involve multiple recommendation scenarios. To determine the recommended object to be displayed to the triggering user from the candidate distribution objects corresponding to the map page, the corresponding triggering scenario is the map page scenario. Therefore, the target triggering scenario can include the map page scenario and the triggering time and space.

[0070] Among them, triggering time and space refers to the triggering conditions under the time and space dimensions. For example, the triggering time and space include triggering time and / or triggering space. The triggering time can refer to time period information, including dimensions such as week, holidays, hours, time periods, etc., such as "Monday", "Saturday", "Sunday", "Spring Festival", "National Day", "Mid-Autumn Festival", "10 am", "3 pm", "9 pm", "dinner time", "lunch time", "working hours", etc.

[0071] The trigger space refers to the trigger location and trigger status of the event displayed on the map page. The trigger location can include dimensions such as city, region, business district, AOI (Area of ​​Interest), Geohash (an algorithm used to encode geographic location (latitude and longitude coordinates) as a short string), and distance. Business districts and AOIs can include current, nearby, frequently visited by the user, and near the user's permanent residence, such as "City A," "Region Z," "Mall X," and "Within 2 kilometers." Trigger statuses can include normal, business trip, tourism, and returning home.

[0072] Specifically, each candidate distribution object is each content to be distributed that is pre-configured in the map page. For example, the candidate distribution object may include page function items, interest objects and / or map display parameters. For example, page function items may include "shopping and eating", "afternoon tea", "movie watching", "attractions", "Sichuan cuisine", etc.; interest objects may include location points (location markers), note points (note markers), group chat points (group chat markers), live broadcast points (live broadcast markers), etc.; map display parameters may refer to scales, display areas, etc. Different triggering scenarios may correspond to different map display parameters. For example, on Monday, at home, at work, and in the same city, the scale is 3 kilometers around the user's location; on holidays, Hotel A, traveling, and in other places, the scale is 20 kilometers around.

[0073] In actual implementation, the relevant historical behavior data of the triggering user can be obtained. The relevant historical behavior data can indicate the historical distribution objects that the triggering user has interacted with and / or not interacted with. The target historical behavior data that meets the target triggering scenario can be filtered out from the relevant historical behavior data to represent the user's historical interest preferences in the target triggering scenario.

[0074] For example, assuming that the target trigger scenario is a map page scenario, the trigger time is "3 p.m.", the trigger location is "Mall A", and the trigger status is "business trip + out of town", at this time, the relevant behavior data of the historical distribution objects that the trigger user has interacted with under scenario conditions such as "3 p.m.", "Mall A", "business trip + out of town" before the current time can be filtered out from the relevant historical behavior data of the trigger user to obtain the target historical behavior data, which indicates the historical preferences of the trigger user under the trigger conditions of "3 p.m.", "Mall A", and "business trip + out of town".

[0075] In the embodiments of this specification, the target triggering scenario includes the map page scenario and the triggering time and space. The target historical behavior data that meets the map page scenario and the triggering time and space can be filtered out from the relevant historical behavior data of the triggering user, so that the map page scenario and the triggering time and space are used as triggering conditions, and the historical preferences of the triggering user under the triggering conditions are determined, and then the recommended objects in the map page are determined to achieve targeted recommendations.

[0076] Step 106: Determine the recommended object under the target triggering scenario from the candidate distribution objects corresponding to the map page according to the target triggering scenario and / or the target historical behavior data.

[0077] Specifically, the candidate distribution objects include page function items, interest objects and / or map display parameters, and the recommended objects in the target trigger scenario filtered out from the candidate objects may include recommended function items, recommended interest objects and / or recommended map display parameters.

[0078] It should be noted that when a user triggers a map page display event and enters the map page, the purpose is to obtain information under clear triggering conditions of time, location, status and other dimensions. Unlike the recommendation scenario where users passively accept input and the search scenario where users need to actively search for information, the map page scenario is a scenario with clear user intentions. Based on the strong needs of users, the content that the user is most interested in at the current time, within the current location range and in the current state can be recommended to the user in combination with the triggering time, triggering location and / or triggering status of the map page display event.

[0079] Therefore, in an embodiment of the present specification, in response to a map page display event, it is necessary to determine the target trigger scenario of the map page display event, obtain the target historical behavior data under the target trigger scenario, and filter out recommended objects from the various candidate distribution objects of the map page to estimate the points that the user may be interested in under a clear trigger time, trigger location, and trigger state.

[0080] For example, Figure 4 This is a schematic diagram of a process for determining a recommended object provided by an embodiment of this specification. Figure 4 As shown, when determining the recommended objects in the map page, it is first necessary to understand the target triggering scenario that triggers the map page display event. For example, the target triggering scenario may include time, location, status, etc. In addition, it is also necessary to understand user behavior, which may include basic information, travel interests, and historical preferences under the target triggering scenario, etc. Furthermore, it is also necessary to understand the distribution content, which includes various candidate distribution objects related to the map page, such as functional items, points of interest (POI), scales, etc. Points of interest include restaurants, attractions, services, group chats, etc. Based on the above content, recommended objects that the user may be interested in under triggering conditions such as time, location, and status can be filtered out from various candidate distribution objects for distribution.

[0081] As an example, the recommended objects determined are targeted recommendations based on the target trigger scenario that triggers the map page display event. Depending on the scenario in which the user triggers the map page display event, the recommended objects determined will differ. For example, when a user accesses the map page at different times, locations, exploration states, and travel statuses, they will prioritize recommended objects that interest them based on the current trigger conditions. The recommended objects corresponding to different times, locations, exploration states, and travel statuses are shown in Table 1 below.

[0082] Table 1 Recommended objects corresponding to different time, different location, different exploration status, and different user travel status

[0083]

[0084]

[0085] In an optional implementation of this embodiment, determining a recommended object under the target triggering scenario from candidate distribution objects corresponding to the map page based on the target triggering scenario and / or target historical behavior data includes:

[0086] Determining, based on the target historical behavior data, a scenario interest vector of the triggering user in the target triggering scenario;

[0087] Obtaining basic features of the triggering user and basic features of the candidate distribution objects;

[0088] Obtaining, through a target recommendation model, a trigger probability of each candidate distribution object in at least one set prediction dimension based on the basic features of the triggering user and the basic features of the candidate distribution objects, the target triggering scenario and / or the scenario interest vector;

[0089] According to the trigger probability, the recommended object for the triggering user in the target triggering scenario is determined.

[0090] In actual implementation, the target historical behavior data includes historical distribution objects that the triggering user has interacted with and / or not interacted with when opening the map page in the target triggering scenario. Therefore, the target historical behavior data can be analyzed to obtain the triggering user's scenario interest vector in the target triggering scenario. The scenario interest vector can indicate the triggering user's interest preferences. For example, if the target triggering scenario is: company + 2 pm + opening the map page, the target historical behavior data can be understood as the map page historically opened at 2 pm in the company, which includes the historical distribution objects that the user has interacted with and / or not interacted with when opening the map page at 2 pm in the company every day in history.

[0091] Specifically, the target's historical behavior data can be analyzed using an interest analysis model to determine the user's scene interest vector in the target triggering scenario. The interest analysis model can be a neural network ranking model (Recurrent Attention over Contextualized Page sequence, RACP). The RACP model is a ranking model that combines a recurrent neural network (RNN) and an attention mechanism. It can be used to process the ranking task of serialized page data (i.e., target historical behavior data). The RACP model can use page context-aware attention to model the context within the map page, allowing for a more specific understanding of user preferences.

[0092] That is to say, the target historical behavior data of the triggering user in the target triggering scenario can be input into the neural network ranking model (RACP) to obtain the user's scenario interest vector. The RACP model can include an intra-page context-aware interest layer, an inter-page interest backtracking layer, and a page-level interest aggregation layer. The intra-page context-aware interest layer extracts feedback information and context information for each page_i (the map page opened in the target historical behavior data) to extract the user interest in each map page; the inter-page interest backtracking layer also stimulates the evolution of users' interests and interactions when browsing different historical candidate distribution objects, because the interest in the previous page will affect the user's interaction with the next page. The user's interest can be inferred from the user's interaction information on the next page, which is called interest backtracking, ensuring the consistency of long-term and short-term interests; the page-level interest aggregation layer is an attention module that can capture the different importance of different map pages to the final prediction. The final user behavior representation is also a high-level aggregation of all contextualized page representations, aggregating the interests of all map pages to obtain the scenario interest vector that triggers the user in the target triggering scenario.

[0093] Of course, in actual implementation, in addition to the above-mentioned RACP model, the interest analysis model can also use other models or tools to analyze the target historical behavior data to obtain the scene interest vector of the triggered user in the target triggering scenario, such as the sequence ranking model based on the attention mechanism (Transformer-based Rankers, Hierarchical Attention Networks, etc.), the hybrid model of recurrent neural network + attention (Neural Attentive Recommendation Machine), etc. This manual does not limit this.

[0094] It should be noted that the basic features include at least one of the following: identification features, attribute features, and interaction features. That is, the basic features of the triggering user refer to at least one of the identification features, attribute features, and interaction features of the triggering user, and the basic features of the candidate distribution object refer to at least one of the identification features, attribute features, and interaction features of the candidate distribution object. Among them, the basic features of the triggering user may include the triggering user identification, triggering user attributes, triggering user interaction features, etc. The triggering user identification is used to identify the triggering user on the content application platform, such as the user ID; the triggering user attributes are some reference information of the triggering user, such as the client version, age, gender, and other features of the client used by the triggering user; the triggering user interaction features are the distribution of the triggering user's behavior within a set time period (such as the same day, within 7 days) (including the triggering user's behavior of clicking notes, purchasing items, searching for content, etc. in various scenarios), such as the triggering user's click / like and other interaction features on the same day, in the past 3 days, and in the past 7 days. The basic features of the candidate distribution objects may include object identification, object attributes, object interaction features, etc. The object identification is used to identify the object on the content application platform, such as the object ID; the object attributes are some reference information, such as categories (first-level categories, second-level categories); the object's interaction features, that is, the distribution of the object's interaction features within a set time period (such as the same day, within 7 days) (such as the object's exposure, forwarding volume, collection volume, and other feature distributions in various scenarios within the set time period), such as the object's click / like and other interaction features on the same day, in the past 3 days, and in the past 7 days.

[0095] Specifically, setting a prediction dimension refers to a prediction dimension for screening recommended objects, such as prediction tasks such as likes, clicks, and comments. For example, if the prediction dimension is set to click-through rate prediction, the target recommendation model can be used to predict the click probability of each candidate distribution object, and the candidate distribution object with a high click probability can be selected as the recommended object; for another example, if the prediction dimension is set to like rate prediction, the target recommendation model can be used to predict the like probability of each candidate distribution object, and the candidate distribution object with a high like probability can be selected as the recommended object; for another example, if the prediction dimension is set to comment rate prediction, the target recommendation model can be used to predict the comment probability of each candidate distribution object, and the candidate distribution object with a high comment probability can be selected as the recommended object.

[0096] It should be noted that the target recommendation model is obtained by offline training of sample behaviors of one or more triggering scenarios. With the help of the target recommendation model, the basic characteristics of the triggering user and the basic characteristics of the candidate distribution objects, the target triggering scenarios and / or scenario interest vectors are analyzed, and the triggering probability of each candidate distribution object in at least one set prediction dimension is predicted. Then, the candidate distribution objects are sorted according to the triggering probability, and a set number of candidate distribution objects with the highest ranking are selected as recommended objects and displayed on the map page.

[0097] In an embodiment of the present specification, the target historical behavior data can be sorted and analyzed to obtain the interest preferences of the triggering user in the target triggering scenario, and the basic characteristics of the triggering user and the basic characteristics of the candidate distribution objects can be obtained. By using the trained target recommendation model, combined with the basic characteristics of the triggering user and the basic characteristics of the candidate distribution objects, the target triggering scenario and / or the scenario interest vector, the triggering probability of each candidate distribution object in at least one set prediction dimension is obtained, the triggering probabilities are sorted, and the objects with higher triggering probabilities are selected as the recommended objects in the target triggering scenario. Based on the user's historical preferences in the target triggering scenario, the basic characteristics of the triggering user, the basic characteristics of the candidate distribution objects, the target triggering scenario and other multi-dimensional information, the objects with higher triggering probabilities are recommended to the user, thereby improving the click-through conversion rate of the recommended objects.

[0098] In an optional implementation of this embodiment, the target recommendation model includes a personalized embedding network, a parameter personalized neural network corresponding to at least one set prediction dimension, and at least one prediction network corresponding to at least one set prediction dimension; obtaining, through the target recommendation model, a trigger probability of each candidate distribution object in at least one set prediction dimension based on the basic characteristics of the triggering user and the basic characteristics of the candidate distribution object, the target triggering scenario and / or the scenario interest vector, includes:

[0099] Based on the basic features of the triggering user and the basic features of the candidate distribution objects, determining a scene personalized embedding vector corresponding to the target triggering scene through the personalized embedding network;

[0100] Based on the target triggering scenario, the scenario personalized embedding vector, and / or the scenario interest vector, obtaining, through the parameter personalized neural network, at least one prediction vector of the triggering user for the candidate distribution object in at least one set prediction dimension under the target triggering scenario;

[0101] Based on the at least one prediction vector, a triggering probability of the triggering user for each candidate distribution object in at least one set prediction dimension is obtained through the at least one prediction network.

[0102] Specifically, the personalized embedding network is used to align the feature importance of each scenario, and the parameter-personalized neural network is used to fine-tune the parameters of the deep neural network (prediction backbone network) to balance prediction preferences across different prediction dimensions. Each prediction network is used to perform predictive analysis on each prediction vector to determine the probability of triggering the user for each candidate distribution object in each set prediction dimension.

[0103] In actual implementation, in different scenarios, the user behavior and the distribution of candidate distribution objects are different. Based on the basic features of the triggering user and the basic features of the candidate distribution objects, the scene personalized embedding vector corresponding to the target triggering scenario can be determined through a personalized embedding network to obtain an embedding vector that incorporates the target triggering scenario information. The basic features of the input triggering user and / or the basic features of the candidate distribution objects are personalized embedded to align the feature importance in different scenarios and avoid the occurrence of a "scene seesaw".

[0104] In addition, the multi-task learning approach uses complex modules to model the representations of multiple set prediction dimensions, that is, stacked deep neural network layers. However, since the parameters of the deep neural network are shared by all set prediction dimensions and lack personalized parameters, the prediction preferences of different set prediction dimensions are not unified, which makes it difficult to balance the multiple set prediction dimensions in modeling. Therefore, based on the target triggering scenario, the scenario personalized embedding vector, and / or the scenario interest vector, a parameter personalized neural network can be used to obtain at least one prediction vector of the candidate distribution object in at least one set prediction dimension of the triggered user in the target triggering scenario. The parameters of different deep neural networks in the parameter personalized neural network are personalized and fine-tuned in each set prediction dimension to balance the prediction preferences of different set prediction dimensions and predict the prediction vector of each candidate distribution object.

[0105] In the embodiments of this specification, the prediction accuracy of the target recommendation model is ensured by aligning the feature importance in different scenarios and balancing the prediction preferences of different set prediction dimensions through personalized embedding networks and parameter personalized neural networks.

[0106] In an optional implementation of this embodiment, the personalized embedding network includes a first gating network; and determining, through the personalized embedding network, the scene personalized embedding vector corresponding to the target triggering scene based on the basic features of the triggering user and the basic features of the candidate distribution objects includes:

[0107] Obtaining statistical features of the target triggering scenario;

[0108] Inputting the statistical features of the target triggering scenario into the first gating network to generate a first personalized gating result;

[0109] According to the first personalized gating result and the basic embedding vector, a scene personalized embedding vector corresponding to the target triggering scene is obtained, wherein the basic embedding vector is obtained based on the basic features of the triggering user and the basic features of the candidate distribution object.

[0110] In actual implementation, the target recommendation model includes an embedding layer, which is used to map the input discrete features into low-dimensional, dense, continuous vectors (i.e., embedding vectors), thereby facilitating model calculation and feature learning. Specifically, the basic features of the triggering user and the basic features of the candidate distribution objects can be input into the embedding layer of the target recommendation model to obtain the feature vectors of the triggering user and the candidate distribution objects, which are then used as the basic embedding vectors.

[0111] It should be noted that the basic features such as the identification features, attribute features, interaction features of the triggering user and the identification features, attribute features, interaction features of the candidate distribution objects can be obtained, and embedded coding can be performed through the embedding layer of the target recommendation model to obtain the feature vector of the triggering user and the feature vector of the candidate distribution object as the basic embedding vector, so as to obtain the scenario-personalized embedding vector.

[0112] In addition, for common basic features such as the basic features of triggering users and the basic features of candidate distribution objects, statistical features of target triggering scenarios, and other discrete features that need to be embedded and encoded, the amount of data is relatively large. In the target recommendation model, an embedding layer is often shared. However, for multiple scenarios, the shared embedding layer ignores the differences between scenarios. Therefore, scene-side features (that is, statistical features of target triggering scenarios) can be introduced to personalize the differences between different scenarios.

[0113] In actual implementation, the personalized embedding network includes a first gating network. The gating network (GateNeuralUnit, GateNU) is the basic unit of the personalized embedding network and the parameter personalized neural network. It is a gating structure that can process more prior knowledge with personalized semantics and map it into the model. In the recommendation scenario, the candidate distribution object side is also very critical, such as object ID, object category, publisher, etc. Users will show different preference patterns for different candidate distribution objects. The gating network can include two layers. The first layer of the network is used for feature intersection of various prior knowledge, and the second layer is used to generate gating scores. The Sigmoid and hyperparameters in the second layer can control the value range of the score vector within the set range.

[0114] Specifically, the statistical features of the target triggering scene can be obtained. The statistical features are a type of scene-side information, such as scene identification, user behavior statistics in the scene, and exposure statistics of each object. The statistical features of the target triggering scene are input into the first gating network to generate a first personalized gating result. The first personalized gating result can be a gating score. The gating score is subjected to a pixel-by-pixel dot product operation with the basic embedding vector (i.e., the feature vector of the triggering user and the feature vector of the candidate distribution object) to obtain a scene-personalized embedding vector corresponding to the target triggering scene.

[0115] In an embodiment of the present specification, a personalized embedding network adds scene-specific personalized information to the basic feature vector input by the embedding layer to generate a personalized gating result, so as to generate a scene-personalized embedding vector. Without changing the structure of the embedding layer, the basic feature vectors such as the feature vector of the triggering user and the feature vector of the candidate distribution object obtained by encoding the embedding layer are personalized converted through the first gating network to align the feature importance of multiple users in different scenarios, ensure the interaction between the features of different users and different candidate distribution objects, and enhance personalization.

[0116] In an optional implementation of this embodiment, the parameter-personalized neural network includes a second gating network and at least one deep neural network corresponding to at least one set prediction dimension; obtaining, through the parameter-personalized neural network, at least one prediction vector of the triggered user for the candidate distribution object in the target triggering scenario in at least one set prediction dimension based on the target triggering scenario, the scenario-personalized embedding vector, and / or the scenario interest vector, includes:

[0117] Inputting the scene interest vector, the scene personalized embedding vector, and the target triggering scene into a second gating network to generate a second personalized gating result;

[0118] The scene interest vector and the scene personalized embedding vector are respectively input into the at least one deep neural network, and the second personalized gating result is applied to each neural network layer of the deep neural network to obtain a prediction vector output by the deep neural network.

[0119] It should be noted that the parameter-personalized neural network includes a second gating unit and a deep neural network. The second gating unit has the same structure as the first gating unit and will not be described in detail here. The deep neural network (DNN) is the backbone prediction network in the target recommendation model. The deep neural network can be composed of multiple neural network layers. Different deep neural networks perform prediction tasks of different prediction dimensions and generate prediction vectors of different prediction dimensions.

[0120] In actual implementation, the prediction tasks of different prediction dimensions have unique sparsity and influence each other. The scene interest vector, scene personalized embedding vector, target trigger scene, etc. can be spliced ​​and input into the second gating network to generate a second personalized gating result. The target trigger scene can provide scene-side reference information. The target trigger scene is used as a priori personalized feature on the scene side. These prior personalized features are spliced ​​with the scene personalized embedding vector output by the personalized embedding network, and the scene interest vector representing the historical interest preference of the triggered user, and input into the second gating network to obtain the second personalized gating result. The second personalized gating result is respectively subjected to pixel-by-pixel dot product operation with the vectors output by each network layer of the deep neural network to amplify or reduce the contribution of the network layer, thereby realizing parameter personalization of the deep neural network.

[0121] Specifically, the first L-1 layers of the deep neural network use Relu (activation function), and the last layer uses Sigmoid (activation function). The output of the last layer is the target personalized vector after parameter personalization, that is, the corresponding prediction vector is output for the set prediction dimension.

[0122] In the embodiments of this specification, the parameter-personalized neural network can optimize the parameters of the deep neural network to balance the different sparsities of different prediction targets (i.e., different set prediction dimensions).

[0123] In an optional implementation of this embodiment, obtaining, based on the at least one prediction vector, through the at least one prediction network, a triggering probability of the triggering user for each candidate distribution object in at least one set prediction dimension includes:

[0124] Obtaining a graph feature vector of the triggering user and / or a graph feature vector of a candidate distribution object, wherein the graph feature vector is obtained based on an interaction graph network, and the interaction graph network is used to indicate interaction relationships between multiple users and multiple objects;

[0125] Based on at least one of the prediction vectors, the graph feature vector of the triggering user and / or the graph feature vector of the candidate distribution object, the triggering probability of the triggering user for each candidate distribution object under the at least one set prediction dimension is obtained through a prediction network corresponding to at least one set prediction dimension.

[0126] In actual implementation, the interaction graph network of the content application platform refers to the full interactive behavior of the content application platform. The interaction graph network is independent of the triggering scenario and can indicate the interactive relationship between multiple users and multiple objects in the content application platform, such as which user clicked on which object. Specifically, the interactive behavior data of each content application platform (including browsing, clicking, interaction and other behavioral data of each user) can be collected. Based on the pre-trained GCN graph neural network, the interactive relationship between multiple users and multiple objects is analyzed to generate a corresponding interaction graph network. The interaction graph network can represent users and objects through nodes, and the connecting edges between nodes indicate whether there is an interactive relationship between users and objects. The characteristics of each node in the interaction graph network are 32-dimensional vectors, which can reflect the characteristics of the corresponding node and the interactive relationship with other nodes.

[0127] It should be noted that, since the interaction graph network can indicate the interaction relationship between multiple users and multiple objects, the feature information of the graph is extracted based on the interaction graph network through specific algorithms and models. For the triggering user, its interaction pattern, connection strength and other information with other objects and users can be analyzed from the interaction graph network, and then the graph feature vector corresponding to the triggering user can be obtained. The graph feature vector can be used to describe the characteristics and attributes of the triggering user in the entire interaction network; for the candidate distribution object, its association with each user can also be analyzed from the interaction graph network, so as to obtain the graph feature vector of the candidate distribution object to reflect the characteristics and status of the object in the interaction network.

[0128] Specifically, the graph feature vector of the triggering user and the graph feature vector of the candidate distribution object can be obtained based on the interactive graph network. For any prediction vector of a set prediction dimension, based on the prediction vector, the graph feature vector of the triggering user, and the graph feature vector of the candidate distribution object, analysis is performed through the prediction network of the set prediction dimension to obtain the triggering probability of the triggering user for each candidate distribution object under the set prediction dimension.

[0129] In the embodiments of this specification, for each set prediction dimension, based on the prediction vector of each set prediction dimension, combined with the graph feature vector of the triggering user and / or the graph feature vector of the candidate distribution object, the prediction network corresponding to each set prediction dimension can be used to analyze and determine the triggering probability of each candidate distribution object for the triggering user under each set prediction dimension. This is actually a multi-scenario multi-task learning, which analyzes information of multiple dimensions to capture the user's preference for different candidate distribution objects in different scenarios, and realizes the triggering probability prediction of each candidate distribution object under each set prediction dimension, so as to select the candidate distribution object that is most likely to attract the user as the recommended object, ensuring that the recommended object is in line with the user's past interests and adapts to the current triggering scenario. With the help of rich behavioral information, the prediction accuracy of the triggering probability is improved, thereby improving the recommendation accuracy of subsequent recommended objects.

[0130] For example, Figure 5 This is a schematic diagram of a trigger probability prediction process based on a target recommendation model provided by an embodiment of this specification. Figure 5 As shown, the target historical behavior data of the triggering user in the target triggering scenario is obtained. The target historical behavior data is the map page opened by the triggering user in the target triggering scenario. The map page includes the historical distribution objects that have been interacted (indicated by "√" in the figure) and the historical distribution objects that have not been interacted (indicated by "×" in the figure). The target historical behavior data is input into the neural network ranking model (RACP) to obtain the scene interest vector of the triggering user in the target triggering scenario.

[0131] Obtain the basic features of the triggering user (SF(1), SF(2), ... SF(n)) and the basic features of the candidate distribution objects (DF(1), DF(2), ... DF(n)), input the basic features of the triggering user and the basic features of the candidate distribution objects into the embedding layer (Embedding Layer) to obtain the basic embedding vector, which includes the feature vector of the triggering user (SF(1'), SF(2'), ... SF(n')) and the feature vector of the candidate distribution object (DF(1'), DF(2'), ... DF(n')).

[0132] Obtain the statistical features of the target triggering scenario (domain-side features) and input them into the embedding layer to obtain a scenario embedding vector. This scenario embedding vector is input into the first gating network (Gate NU) to generate a first personalized gating result. This first personalized gating result is dot-producted with the feature vector of the triggering user and the feature vector of the candidate distribution object, and then concatenated to obtain a personalized scenario embedding vector.

[0133] The target triggering scene (time, location, scene) is input into the embedding layer to obtain the encoding features of the target triggering scene. The scene interest vector, the scene personalized embedding vector, and the encoding features of the target triggering scene are spliced ​​and input into the second gating network to obtain the second personalized gating result. In addition, the scene interest vector and the scene personalized embedding vector can be spliced ​​and input into the deep neural network to obtain the second personalized gating result. Figure 5 The example shown includes two deep neural networks, which are used to output prediction vectors for two different set prediction dimensions. The deep neural network includes four neural network layers. The output vector of each neural network layer is dot-producted with the second personalized gating result. The output of the last layer is the prediction vector after parameter personalization for the corresponding set prediction dimension. The graph feature vector of the triggering user and / or the graph feature vector of the candidate distribution object (obtained based on the interactive graph network) and the prediction vector are input into the corresponding prediction network. The corresponding set prediction dimension is predicted through the prediction network to obtain the triggering probability of each candidate distribution object under the corresponding set prediction dimension.

[0134] in, Figure 5 “⊕” in the figure indicates the splicing operation; Represents the dot product operation; Indicates that gradient backpropagation is canceled.

[0135] In an optional implementation of this embodiment, the target recommendation model is trained by the following method:

[0136] Obtaining sample behavior data for one or more trigger scenarios, wherein the sample behavior data includes a sample user, a sample distribution object, and a trigger tag in a set prediction dimension, wherein the trigger tag indicates an interaction relationship between the sample user and the sample distribution object in the set prediction dimension under the corresponding trigger scenario;

[0137] Based on the sample behavior data of the one or more triggering scenarios, the initial recommendation model is trained to obtain a trained target recommendation model.

[0138] Among them, sample behavior data refers to the operation data and sample distribution objects of a large number of sample users obtained offline, which serve as training samples for the target recommendation model, and the trigger label serves as the real label under the set prediction dimension.

[0139] In actual implementation, the sample users and sample distribution objects in the sample behavior data of one or more trigger scenarios can be used as input data, and the trigger labels can be used as sample labels for setting prediction dimensions. The initial recommendation model can be supervised trained to obtain the trained target recommendation model.

[0140] Specifically, sample users and sample distribution objects in one or more trigger scenarios are input into the initial recommendation model to obtain the predicted probability of each candidate distribution object predicted by the initial recommendation model in the set prediction dimension for each trigger scenario, and the predicted trigger object is determined based on the predicted probability. According to the predicted trigger object and the corresponding trigger label in each trigger scenario, the corresponding loss value is calculated, and the model parameters of the initial recommendation model are adjusted according to the reverse gradient propagation of the loss value. The initial recommendation model is continuously trained until the training stop condition is reached to obtain the target recommendation model after the training is completed.

[0141] In an optional implementation of this embodiment, the gradients calculated by the parameter-based personalized neural network in the target recommendation model during the training phase are not propagated back to the personalized embedding network to avoid affecting the updates of the embedding layer in the personalized embedding network. In other words, during forward prediction, probabilistic predictions can be made based on sample users and sample distribution targets in one or more triggering scenarios, but the parameter-based personalized neural network does not participate in the loss-based backward gradient propagation.

[0142] It's important to note that during the training of the target recommendation model, the parameter-based personalized neural network calculates gradients, but with a special setting, the calculated gradients are not propagated back to the personalized embedding network. Gradient propagation is typically a crucial step in neural network training for updating network parameters. Gradient information is used to adjust the weights of individual neurons in the network, bringing the model's predictions closer to the true values. However, the reason for not propagating gradients back to the personalized embedding network is to avoid affecting parameter updates in the embedding layer within the personalized embedding network. The embedding layer converts high-dimensional, discrete data into a low-dimensional, continuous vector representation. Its parameters are crucial for the model to learn the characteristics and patterns of the data, and therefore, we do not want them to be interfered with by the gradients of the parameter-based personalized neural network.

[0143] In one possible implementation, whether the training stop condition has been met can be determined solely based on the relationship between the loss value and the loss threshold. Specifically, if the loss value is greater than or equal to the loss value threshold, it indicates that the predicted trigger object and the corresponding trigger label are significantly different, and the initial recommendation model's scoring prediction ability is poor. In this case, it can be determined that the training stop condition has not been met, and the model parameters of the initial recommendation model are adjusted. The initial recommendation model is then trained until the loss value is less than the loss value threshold, indicating that the difference between the predicted trigger object and the corresponding trigger label is small. It is then determined that the training stop condition has been met, and training is stopped to obtain the trained target recommendation model.

[0144] The loss value threshold is a critical value of the loss value. If the loss value is greater than or equal to the loss value threshold, it indicates that there is still a certain deviation between the prediction result of the initial recommendation model and the actual result, and the model parameters of the initial recommendation model still need to be adjusted. At this time, it is determined that the training stop condition has not been met. If the loss value is less than the loss value threshold, it indicates that the prediction result of the initial recommendation model is close enough to the actual result, and training can be stopped. At this time, it can be determined that the training stop condition has been met. The numerical value of the loss value threshold is selected according to the actual situation, and the embodiments of the present application do not impose any restrictions on this.

[0145] In another possible implementation, in addition to comparing the relationship between the loss value and the loss value threshold, the number of iterations can also be combined to determine whether the training stop condition has been met. Specifically, if the loss value is greater than the loss value threshold, it can be further determined whether the number of iterations at this moment has reached the preset number of iterations. If the number of iterations at this moment has not reached the preset number of iterations, it can be determined that the training stop condition has not been met. The model parameters of the initial recommendation model can be adjusted, and the initial recommendation model can be continued to be trained until the preset number of iterations is reached. In this case, it is determined that the training stop condition has been met, and the iterations are stopped to obtain the trained target recommendation model.

[0146] Among them, the preset number of iterations is set according to the actual situation, and the embodiment of the present application does not impose any restrictions on this. When the number of training times reaches the preset number of iterations, it means that the number of training times of the initial recommendation model is sufficient. At this time, the prediction result of the initial recommendation model is extremely close to the actual result, and the training can be stopped.

[0147] In practical applications, there are many loss functions used to calculate loss values, such as cross-entropy loss, L1-norm loss, maximum loss, mean squared error, and logarithmic loss. Loss functions can be used to evaluate the degree of discrepancy between a model's predicted results and the actual results. A better loss function generally indicates better model performance, and different types of models generally use different preferred loss functions.

[0148] In an embodiment of the present specification, sample behavior data of one or more trigger scenarios can be obtained to train the initial recommendation model, so that the trained target recommendation model can distinguish different trigger scenarios and predict the trigger probability of each candidate distribution object. Then, with the help of the target recommendation model obtained by training, the trigger probability prediction in a specific trigger scenario can be realized, thereby realizing the screening of recommended objects in a specific trigger scenario.

[0149] Step 108: Distribute the recommended objects to the client, so that the client displays the map page and displays the recommended objects in the target triggering scenario on the map page.

[0150] It should be noted that after the server determines the recommended object under the target trigger scenario, it can distribute the recommended object to the client so that the client can display the map page and display the recommended object on the map page, so that the recommended objects displayed on the map page are different under different trigger scenarios.

[0151] In an embodiment of the present specification, a map page display method is provided. In response to a map page display event, a triggering scenario such as the map page scene, current time, current location, and current state that triggered the map page display event can be obtained. Based on the target historical behavior data under the triggering scenario, with the help of a trained target recommendation model, a trigger probability prediction is performed for each candidate distribution object of the map page, and recommended objects under the current time, current location, and current state under the map page scenario are screened out and distributed to the client for display on the map page. In this way, based on the current time, current location, and current state under the map page scenario, recommended objects to be displayed on the map page can be screened out. When users enter the map page at different times, locations, and states, different recommended objects are displayed. Recommended objects that users may be interested in under triggering scenarios such as the current time, current location, and current state can be recommended to users, thereby achieving personalized recommendations that better meet user needs, facilitate user operation, and improve user experience, thereby improving the conversion rate of recommended objects on the map page and increasing user stickiness.

[0152] See also Figure 6 , Figure 6 A flowchart of another map page display method provided according to an embodiment of the present specification is shown, which is applied to a client and specifically includes the following steps.

[0153] Step 602: In response to a map page display event, obtain a recommended object corresponding to the target trigger scenario of the map page display event, wherein the recommended object is determined from the candidate distribution objects corresponding to the map page based on the target trigger scenario and / or the target historical behavior data under the target trigger scenario.

[0154] It should be noted that in response to the map page display event, it indicates that the user wants to enter the map page. At this time, the recommended object corresponding to the target triggering scene of the map page display event can be obtained. The recommended object corresponding to the target triggering scene can be obtained through the above Figure 1 The contents described in the illustrated embodiment are confirmed, and the embodiments of this specification will not be repeated here.

[0155] Step 604: Display the map page and display the recommended objects on the map page.

[0156] In an optional implementation of this embodiment, the recommended objects include page function items; and the recommended objects are displayed on the map page, including:

[0157] Display the page function items corresponding to the target trigger scenario on the map page;

[0158] In response to the triggering operation of the target function item, a target point corresponding to the target function item is acquired, and the target point corresponding to the target function item is displayed on a map page.

[0159] It should be noted that the recommended objects include page function items, that is, when the user enters the map page under different triggering scenarios, the page function items displayed are different.

[0160] In actual implementation, a map page can be displayed in response to a map page display event, and page function items corresponding to the target trigger scenario can be displayed on the map page. The user can trigger any target function item, and the target point corresponding to the target function item can be filtered out from various points and displayed on the map page, filtering out other recommended points outside the target function item.

[0161] Specifically, in response to the triggering operation of the target function item, the client can send a data acquisition request to the server, and the server will return the target point of the target function item; or, if the client has obtained the recommended points from the server, the client can also analyze the attribute category of each recommended point based on the identification of the target function item, and filter out the recommended points whose attribute category matches the target function item as the target point.

[0162] For example, Figure 7 This is a schematic diagram of a first map page provided by an embodiment of this specification, such as Figure 7 For example, suppose a user triggers a map page display event under the trigger conditions of "3 p.m.," "Mall A," and "Business trip + out-of-town." The recommended function items obtained are "Shopping and Dining" and "Afternoon Tea." Therefore, the map page can display the fixed function items "Pets," "Play," and "Dining." Furthermore, the page function items "Shopping and Dining" and "Afternoon Tea" under the trigger conditions of "3 p.m.," "Mall A," and "Business trip + out-of-town" can be displayed. If the user clicks "Shopping and Dining," the corresponding target points for "Shopping and Dining" can be displayed on the map page: "M Hot Pot," "KK Stir-fry," "BB Dessert," "YY Cafe," "H Photo Spot," "ZZ Cafe," "UU Banquet Hall," and "XX Jewelry Store."

[0163] Figure 8 This is a schematic diagram of a second map page provided in one embodiment of this specification, such as Figure 8For example, suppose a user triggers a map page display event under the trigger conditions of "7 PM," "Mall A," and "Business trip + out-of-town." The recommended function items obtained are "Dinner" and "Movie." Therefore, the map page can display the fixed function items "Pets," "Play," and "Dinner." Also, the page function items "Night Tour" and "Movie" under the trigger conditions of "7 PM," "Mall A," and "Business trip + out-of-town" can be displayed. If the user clicks "Night Tour," the map page can display the corresponding target points for "Night Tour," including "City Park," "P Bookstore," "H Photo Spot," "WW Plaza," "XX Jewelry Store," "YY Cafe," "XX Art Bakery," and "CC Sports Center."

[0164] In the embodiments of this specification, the page function items of the map page can also display different page function items based on different scene trigger conditions, so that the user enters the map page under different time, location, status and other trigger scenarios, and the displayed page function items are different. The page function items that the user is most likely to be interested in under the current trigger scenario are displayed to the user, thereby realizing personalized display of the page function items. After selecting any target function item, the target point corresponding to the target function item can be filtered out for display, and the points that the user is not interested in can be filtered out, which facilitates user operation and improves the user experience.

[0165] In an optional implementation of this embodiment, the recommended objects include objects of interest; and displaying the recommended objects on the map page includes:

[0166] The object of interest corresponding to the target triggering scenario is displayed on the map page, wherein the object of interest displays object information and published content information.

[0167] In actual implementation, recommended objects include objects of interest. Objects of interest corresponding to the target trigger scenario issued by the server can be displayed on the map page. These objects of interest display object information and published content information, making it easier for users to understand the information of each object of interest displayed on the map page. Object information can include information such as the object's icon and name; published content information includes information about published content related to the object of interest, such as the number of published content and whether the location has been visited before. If the published content is a note, the published content information includes the number of notes, etc.

[0168] For example, Figure 9 This is a schematic diagram of a third type of map page provided in one embodiment of this specification, such as Figure 9 Assume that the user triggers the map page display event under the trigger conditions of "3 pm", "A shopping mall", and "business trip + out of town", and the objects of interest obtained are: "XX jewelry store", "P bookstore", "BB dessert", "YY coffee shop", "H photo spot", "history museum", "XX group chat", "city park", "ZZ coffee shop", "XX art bakery", and "CC sports hall".

[0169] In an optional implementation of this embodiment, the recommended objects include objects of interest and map display parameters; displaying the recommended objects on the map page includes:

[0170] The map page is displayed based on the map display parameters, and the object of interest is displayed on the map page.

[0171] It should be noted that, based on triggering scenarios such as the current time, current location, and current status of the map page display event, corresponding map display parameters, such as scale, and corresponding objects of interest under the map display parameters can be filtered out.

[0172] As an example, assuming that the target trigger scenario of the map page display event is "A District, City A", "7 pm", "Normal + Same City", it means that the user is likely to browse nearby restaurants on the map page and needs to know the information and location of nearby restaurants. Therefore, the recommended scale can be determined to be "within 2 kilometers". At this time, a map of the area within 2 kilometers centered on the user's current location can be displayed on the map page, and objects of interest within the 2 kilometers (mainly including recommended restaurants) can be displayed on the map.

[0173] As another example, suppose the target trigger scenario of the event displayed on the map page is "City B", "3 pm", "Travel + Same City", which means that the user is likely to need to browse most of the popular attractions in City B on the map page. Therefore, the recommended scale can be determined to be "20 kilometers". At this time, a map of the area within 20 kilometers centered on the user's current location can be displayed on the map page, and objects of interest within the area within 20 kilometers (mainly including recommended attractions) can be displayed on the map.

[0174] It should be noted that the map page display parameters (such as the scale of the map page) can also display different map pages based on different scene trigger conditions, so that users can enter the map page under different time, location, status and other trigger scenarios. The map page display parameters are different, and the user is shown the map page that may be adapted to their browsing needs under the current trigger scenario, realizing the personalized display of map display parameters, avoiding users from frequently zooming and dragging the map, facilitating user operations, and improving the user experience.

[0175] Of course, in actual implementation, the recommended objects filtered based on the target trigger scenario of the map page display event in the embodiment of this specification may also only include map display parameters. After the map display parameters are determined, the recommended points in the map page are displayed based on the map display parameters (not filtered based on the target trigger scenario). The embodiment of this specification does not limit this.

[0176] A map page display method is provided in an embodiment of the present specification. In response to a map page display event, the triggering scenarios such as the current time, current location, and current status that trigger the map page display event can be obtained, and recommended objects that the user may be interested in under the triggering scenarios such as the current time, current location, and current status can be recommended to the user, thereby achieving personalized recommendations that are more in line with user needs, facilitating user operations, and improving user experience, thereby increasing the conversion rate of recommended objects in the map page and improving user stickiness.

[0177] Figure 10 This is a timing diagram of the processing process of a map page display method provided in an embodiment of this specification, which specifically includes the following steps.

[0178] The client detects the map page display event and obtains the current time, current location, current user travel status, and current exploration status that triggered the map page display event.

[0179] The client sends the current time, current location, current user travel status, and current exploration status to the server.

[0180] On the server side, the target historical behavior data under the triggering scenarios of the current time, current location, current user travel status and current exploration status are obtained. Based on the target historical behavior data and / or the triggering scenarios of the current time, current location, current user travel status and current exploration status, the target recommendation model is called to obtain the triggering probability of each candidate distribution object in at least one set prediction dimension; based on the triggering probability of each candidate distribution object in at least one set prediction dimension, a set number of recommended objects are screened out.

[0181] The server returns the map page to be displayed by the map page display event, as well as the recommended objects under the triggering scenario of the current time, current location, current user travel status and current exploration status to the client.

[0182] The client displays a map page and displays recommended objects in triggering scenarios based on the current time, current location, current user travel status, and current exploration status.

[0183] A map page display method is provided in an embodiment of the present specification. In response to a map page display event, the triggering scenarios such as the current time, current location, and current status that trigger the map page display event can be obtained, and recommended objects that the user may be interested in under the triggering scenarios such as the current time, current location, and current status can be recommended to the user, thereby achieving personalized recommendations that are more in line with user needs, facilitating user operations, and improving user experience, thereby increasing the conversion rate of recommended objects in the map page and improving user stickiness.

[0184] Corresponding to the above method embodiment, this specification also provides a map page display device embodiment, Figure 11A schematic diagram of the structure of a map page display device provided by an embodiment of this specification is shown, which is applied to a server, such as Figure 11 As shown, the device includes:

[0185] The first acquisition module 1102 is configured to, in response to a map page display event, acquire a target triggering scenario corresponding to the map page display event;

[0186] The second acquisition module 1104 is configured to acquire target historical behavior data in a target triggering scenario;

[0187] The determination module 1106 is configured to determine a recommended object under the target triggering scenario from among the candidate distribution objects corresponding to the map page according to the target triggering scenario and / or the target historical behavior data;

[0188] The distribution module 1108 is configured to distribute the recommended objects to the client, so that the client displays the map page and displays the recommended objects in the target triggering scenario on the map page.

[0189] Optionally, the target triggering scene includes a map page scene and a triggering time and space;

[0190] The second acquisition module 1104 is further configured to:

[0191] Obtain relevant historical behavior data of the triggering user;

[0192] Target historical behavior data that meets the target trigger scenario is filtered out from the relevant historical behavior data, and the target historical behavior data includes historical distribution objects that the triggering user has interacted with and / or has not interacted with in the target trigger scenario.

[0193] Optionally, the determination module 1106 is further configured to:

[0194] Determining, based on the target historical behavior data, a scenario interest vector of the triggering user in the target triggering scenario;

[0195] Obtaining basic features of the triggering user and basic features of the candidate distribution objects;

[0196] Obtaining, through a target recommendation model, a trigger probability of each candidate distribution object in at least one set prediction dimension based on the basic features of the triggering user and the basic features of the candidate distribution objects, the target triggering scenario and / or the scenario interest vector;

[0197] According to the trigger probability, the recommended object for the triggering user in the target triggering scenario is determined.

[0198] Optionally, the target recommendation model includes a personalized embedding network, a parameter personalized neural network corresponding to at least one set prediction dimension, and at least one prediction network corresponding to at least one set prediction dimension; the determination module 1106 is further configured to:

[0199] Based on the basic features of the triggering user and the basic features of the candidate distribution objects, determining a scene personalized embedding vector corresponding to the target triggering scene through the personalized embedding network;

[0200] Based on the target triggering scenario, the scenario personalized embedding vector, and / or the scenario interest vector, obtaining, through the parameter personalized neural network, at least one prediction vector of the triggering user for the candidate distribution object in at least one set prediction dimension under the target triggering scenario;

[0201] Based on the at least one prediction vector, a triggering probability of the triggering user for each candidate distribution object in at least one set prediction dimension is obtained through the at least one prediction network.

[0202] Optionally, the personalized embedded network includes a first gated network; the determining module 1106 is further configured to:

[0203] Obtaining statistical features of the target triggering scenario;

[0204] Inputting the statistical features of the target triggering scenario into the first gating network to generate a first personalized gating result;

[0205] According to the first personalized gating result and the basic embedding vector, a scene personalized embedding vector corresponding to the target triggering scene is obtained, wherein the basic embedding vector is obtained based on the basic features of the triggering user and the basic features of the candidate distribution object.

[0206] Optionally, the parameter-personalized neural network includes a second gating network and at least one deep neural network corresponding to at least one set prediction dimension; the determination module 1106 is further configured to:

[0207] Inputting the scene interest vector, the scene personalized embedding vector, and the target triggering scene into a second gating network to generate a second personalized gating result;

[0208] The scene interest vector and the scene personalized embedding vector are respectively input into the at least one deep neural network, and the second personalized gating result is applied to each neural network layer of the deep neural network to obtain a prediction vector output by the deep neural network.

[0209] Optionally, the determination module 1106 is further configured to:

[0210] Obtaining a graph feature vector of the triggering user and / or a graph feature vector of a candidate distribution object, wherein the graph feature vector is obtained based on an interaction graph network, wherein the interaction graph network is used to indicate interaction relationships between multiple users and multiple objects;

[0211] Based on at least one of the prediction vectors, the graph feature vector of the triggering user and / or the graph feature vector of the candidate distribution object, the triggering probability of the triggering user for each candidate distribution object under the at least one set prediction dimension is obtained through a prediction network corresponding to at least one set prediction dimension.

[0212] Optionally, the device further includes a training module configured to:

[0213] Obtaining sample behavior data for one or more trigger scenarios, wherein the sample behavior data includes a sample user, a sample distribution object, and a trigger tag in a set prediction dimension, wherein the trigger tag indicates an interaction relationship between the sample user and the sample distribution object in the set prediction dimension under the corresponding trigger scenario;

[0214] Based on the sample behavior data of the one or more triggering scenarios, the initial recommendation model is trained to obtain a trained target recommendation model.

[0215] Optionally, the gradients calculated by the parameter-personalized neural network during the training phase are not passed back to the personalized embedding network.

[0216] A map page display device is provided in an embodiment of the present specification. It can filter out recommended objects to be displayed on the map page based on the triggering scenario of the map page display event. When the user enters the map page in different triggering scenarios, the recommended objects displayed are different. The recommended objects that the user may be interested in in the current triggering scenario can be recommended to the user, thereby realizing personalized recommendations for different triggering scenarios, better fitting the user's needs in the current triggering scenario, facilitating user operation, improving user experience, and thereby improving the conversion rate of recommended objects in the map page, ensuring user stickiness.

[0217] Corresponding to the above method embodiment, this specification also provides another map page display device embodiment, Figure 12 A schematic diagram of the structure of another map page display device provided by an embodiment of this specification is shown, which is applied to a server, such as Figure 12 As shown, the device includes:

[0218] The third acquisition module 1202 is configured to, in response to a map page display event, acquire a recommended object corresponding to a target triggering scenario of the map page display event, wherein the recommended object is determined from candidate distribution objects corresponding to the map page based on the target triggering scenario and / or the target historical behavior data under the target triggering scenario;

[0219] The display module 1204 is configured to display a map page and display recommended objects on the map page.

[0220] Optionally, the recommended object includes a page function item; the display module 1204 is further configured to:

[0221] Display the page function items corresponding to the target trigger scenario on the map page;

[0222] In response to the triggering operation of the target function item, a target point corresponding to the target function item is acquired, and the target point corresponding to the target function item is displayed on a map page.

[0223] Optionally, the recommended objects include objects of interest; the display module 1204 is further configured to:

[0224] The object of interest corresponding to the target triggering scenario is displayed on the map page, wherein the object of interest displays object information and published content information.

[0225] Optionally, the recommended objects include objects of interest and map display parameters; the display module 1204 is further configured to:

[0226] The map page is displayed based on the map display parameters, and the object of interest is displayed on the map page.

[0227] A map page display device is provided in an embodiment of the present specification. It can filter out recommended objects to be displayed on the map page based on the triggering scenario of the map page display event. When the user enters the map page in different triggering scenarios, the recommended objects displayed are different. The recommended objects that the user may be interested in in the current triggering scenario can be recommended to the user, thereby realizing personalized recommendations for different triggering scenarios, better fitting the user's needs in the current triggering scenario, facilitating user operation, improving user experience, and thereby improving the conversion rate of recommended objects in the map page, ensuring user stickiness.

[0228] The above is a schematic diagram of a map page display device according to this embodiment. It should be noted that the technical solution of this map page display device and the technical solution of the aforementioned map page display method are based on the same concept. For details not described in detail in the technical solution of the map page display device, please refer to the description of the technical solution of the aforementioned map page display method.

[0229] Figure 1313. The block diagram of a computing device according to one embodiment of the present disclosure is shown. Components of the computing device 1300 include, but are not limited to, a memory 1310 and a processor 1320. The processor 1320 is connected to the memory 1310 via a bus 1330, and a database 1350 is used to store data.

[0230] The computing device 1300 also includes an access device 1340 that enables the computing device 1300 to communicate via one or more networks 1360. Examples of these networks include a public switched telephone network (PSTN), a local area network (LAN), a wide area network (WAN), a personal area network (PAN), or a combination of communication networks such as the Internet. The access device 1340 may include one or more of any type of network interface (e.g., a network interface card (NIC)) whether wired or wireless, such as an IEEE 802.11 wireless local area network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a universal serial bus (USB) interface, a cellular network interface, a Bluetooth interface, or a near field communication (NFC) interface.

[0231] In one embodiment of the present specification, the above components of the computing device 1300 and Figure 13 Other components not shown in the figure may also be connected to each other, for example, via a bus. Figure 13 The computing device structure block diagram shown is for illustrative purposes only and is not intended to limit the scope of this specification. Those skilled in the art may add or replace other components as needed.

[0232] Computing device 1300 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, personal digital assistant, laptop computer, notebook computer, netbook computer, etc.), a mobile phone (e.g., a smartphone), a wearable computing device (e.g., a smartwatch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or personal computer (PC). Computing device 1300 may also be a mobile or stationary server.

[0233] The processor 1320 is configured to execute the following computer-executable instructions, which, when executed by the processor, implement the steps of the above-mentioned map page display method.

[0234] The above is a schematic diagram of a computing device according to this embodiment. It should be noted that the technical solution of the computing device and the technical solution of the map page display method described above are based on the same concept. For details not described in detail in the technical solution of the computing device, please refer to the description of the technical solution of the map page display method described above.

[0235] An embodiment of the present specification further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the above-mentioned map page display method.

[0236] The above is a schematic diagram of a computer-readable storage medium according to this embodiment. It should be noted that the technical solution of this storage medium is based on the same concept as the technical solution of the map page display method described above. For details not described in detail in the technical solution of the storage medium, please refer to the description of the technical solution of the map page display method described above.

[0237] An embodiment of the present specification further provides a computer program, wherein when the computer program is executed in a computer, the computer is caused to execute the steps of the above-mentioned map page display method.

[0238] The above is a schematic solution of a computer program of this embodiment. It should be noted that the technical solution of this computer program and the technical solution of the map page display method described above are based on the same concept. For details not described in detail in the technical solution of the computer program, please refer to the description of the technical solution of the map page display method described above.

[0239] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0240] Computer instructions include computer program code, which may be in source code form, object code form, executable files, or some intermediate form. Computer-readable media may include any entity or device capable of carrying computer program code, recording media, USB flash drives, mobile hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunications signals, and software distribution media. It should be noted that the content of computer-readable media may be appropriately increased or decreased based on the requirements of patent practice. For example, in some regions, according to patent practice, computer-readable media does not include electric carrier signals and telecommunications signals.

[0241] It should be noted that for the aforementioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0242] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0243] The preferred embodiments disclosed above are intended only to help illustrate this specification. The optional embodiments do not exhaustively describe all details, nor do they limit the invention to the specific embodiments described. Obviously, many modifications and variations can be made based on the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that those skilled in the art can better understand and utilize this specification. This specification is limited only by the claims and their full scope and equivalents.

Claims

1. A map page display method, characterized in that: Applied to the server, including: In response to a map page display event, obtaining a target triggering scene corresponding to the map page display event; Obtaining historical target behavior data under the target triggering scenario; Determining a recommended object under the target triggering scenario from candidate distribution objects corresponding to the map page according to the target triggering scenario and / or the target historical behavior data; The recommended objects are distributed to a client, so that the client displays the map page and displays the recommended objects in the target triggering scenario on the map page.

2. The map page display method according to claim 1, wherein: The target triggering scene includes a map page scene and a triggering time and space; The acquiring of the target historical behavior data in the target triggering scenario includes: Obtain relevant historical behavior data of the triggering user; Target historical behavior data that meets the target trigger scenario is filtered out from the relevant historical behavior data, and the target historical behavior data includes historical distribution objects that the triggering user has interacted with and / or has not interacted with in the target trigger scenario.

3. The map page display method according to claim 1, wherein: The determining, based on the target triggering scenario and / or the target historical behavior data, a recommended object under the target triggering scenario from each candidate distribution object corresponding to the map page includes: Determining, based on the target historical behavior data, a scenario interest vector of the triggering user in the target triggering scenario; Acquire basic features of the triggering user and basic features of the candidate distribution objects; Obtaining, by means of a target recommendation model, a trigger probability of each candidate distribution object in at least one set prediction dimension based on the basic features of the triggering user and the basic features of the candidate distribution objects, the target triggering scenario and / or the scenario interest vector; According to the trigger probability, a recommended object for the triggering user in the target triggering scenario is determined.

4. The map page display method according to claim 3, wherein: The target recommendation model includes a personalized embedding network, a parameter personalized neural network corresponding to at least one set prediction dimension, and at least one prediction network corresponding to at least one set prediction dimension; obtaining the triggering probability of each candidate distribution object in at least one set prediction dimension through the target recommendation model based on the basic characteristics of the triggering user and the basic characteristics of the candidate distribution objects, the target triggering scenario and / or the scenario interest vector, includes: Based on the basic features of the triggering user and the basic features of the candidate distribution objects, determining a scene personalized embedding vector corresponding to the target triggering scene through the personalized embedding network; Based on the target triggering scenario, the scenario personalized embedding vector, and / or the scenario interest vector, obtaining, through the parameter personalized neural network, at least one prediction vector of the triggering user for the candidate distribution object in at least one set prediction dimension under the target triggering scenario; Based on the at least one prediction vector, a triggering probability of the triggering user for each candidate distribution object in at least one set prediction dimension is obtained through the at least one prediction network.

5. The map page display method according to claim 4, characterized in that: The personalized embedding network includes a first gating network; and determining, through the personalized embedding network, a scene personalized embedding vector corresponding to the target triggering scene based on the basic features of the triggering user and the basic features of the candidate distribution objects, includes: Obtaining statistical features of the target triggering scenario; Inputting the statistical features of the target triggering scenario into the first gating network to generate a first personalized gating result; According to the first personalized gating result and the basic embedding vector, a scene personalized embedding vector corresponding to the target triggering scene is obtained, wherein the basic embedding vector is obtained based on the basic features of the triggering user and the basic features of the candidate distribution object.

6. The map page display method according to claim 4, characterized in that: The parameter-personalized neural network includes a second gating network and at least one deep neural network corresponding to at least one set prediction dimension; obtaining, through the parameter-personalized neural network, at least one prediction vector of the triggered user for the candidate distribution object in the target triggering scenario in at least one set prediction dimension based on the target triggering scenario, the scenario-personalized embedding vector, and / or the scenario interest vector, includes: Inputting the scene interest vector, the scene personalized embedding vector, and the target triggering scene into a second gating network to generate a second personalized gating result; The scene interest vector and the scene personalized embedding vector are respectively input into the at least one deep neural network, and the second personalized gating result is applied to each neural network layer of the deep neural network to obtain a prediction vector output by the deep neural network.

7. The map page display method according to claim 4, characterized in that: The obtaining, based on the at least one prediction vector and through the at least one prediction network, a triggering probability of the triggering user for each candidate distribution object in at least one set prediction dimension includes: Obtaining a graph feature vector of the triggering user and / or a graph feature vector of a candidate distribution object, wherein the graph feature vector is obtained based on an interaction graph network, wherein the interaction graph network is used to indicate interaction relationships between multiple users and multiple objects; Based on at least one of the prediction vectors, the graph feature vector of the triggering user and / or the graph feature vector of the candidate distribution object, the triggering probability of the triggering user for each candidate distribution object under the at least one set prediction dimension is obtained through a prediction network corresponding to at least one set prediction dimension.

8. The map page display method according to claim 4, characterized in that: The target recommendation model is trained by the following method: Obtaining sample behavior data for one or more trigger scenarios, wherein the sample behavior data includes a sample user, a sample distribution object, and a trigger tag in a set prediction dimension, wherein the trigger tag indicates an interaction relationship between the sample user and the sample distribution object in the set prediction dimension under the corresponding trigger scenario; Based on the sample behavior data of the one or more triggering scenarios, the initial recommendation model is trained to obtain a trained target recommendation model.

9. The map page display method according to claim 8, characterized in that: The gradients calculated by the parameter-personalized neural network during the training phase are not passed back to the personalized embedding network.

10. A map page display method, characterized in that: Applied to the client, including: In response to a map page display event, obtaining a recommended object corresponding to a target triggering scenario of the map page display event, wherein the recommended object is determined from candidate distribution objects corresponding to the map page based on the target triggering scenario and / or target historical behavior data under the target triggering scenario; The map page is displayed, and the recommended objects are displayed on the map page.

11. The map page display method according to claim 10, wherein: The recommended object includes a page function item; and displaying the recommended object on the map page includes: Displaying page function items corresponding to the target triggering scenario on the map page; In response to a triggering operation of a target function item, a target point corresponding to the target function item is acquired, and the target point corresponding to the target function item is displayed on the map page.

12. The map page display method according to claim 10, wherein: The recommended objects include objects of interest; and displaying the recommended objects on the map page includes: The object of interest corresponding to the target triggering scenario is displayed on the map page, wherein the object of interest is displayed with object information and published content information.

13. The map page display method according to claim 10, characterized in that: The recommended objects include objects of interest and map display parameters; The displaying of the recommended object on the map page includes: The map page is displayed based on the map display parameters, and the object of interest is displayed on the map page.

14. A map page display device, characterized in that: Applied to the server, including: A first acquisition module is configured to, in response to a map page display event, acquire a target triggering scene corresponding to the map page display event; A second acquisition module is configured to acquire target historical behavior data in the target triggering scenario; a determination module configured to determine, based on the target triggering scenario and / or the target historical behavior data, a recommended object under the target triggering scenario from candidate distribution objects corresponding to the map page, wherein the candidate distribution objects include page function items, objects of interest, and / or map display parameters; The distribution module is configured to distribute the recommended objects to the client, so that the client displays the map page and displays the recommended objects in the target triggering scenario on the map page.

15. A map page display device, characterized in that: Applied to the client, including: a third acquisition module configured to, in response to a map page display event, acquire a recommended object corresponding to a target triggering scenario of the map page display event, wherein the recommended object is determined from candidate distribution objects corresponding to the map page based on the target triggering scenario and / or the target historical behavior data under the target triggering scenario; The display module is configured to display the map page and display the recommended objects on the map page.

16. A computing device, characterized in that include: memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the map page display method according to any one of claims 1 to 13 are implemented.

17. A computer-readable storage medium, characterized in that It stores computer-executable instructions, which, when executed by a processor, implement the steps of the map page display method described in any one of claims 1 to 13.

18. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the map page display method according to any one of claims 1 to 13.

Images