WO2023087915A1 - 应用推荐方法、装置、设备和计算机可读存储介质 - Google Patents

应用推荐方法、装置、设备和计算机可读存储介质 Download PDF

Info

Publication number
WO2023087915A1
WO2023087915A1 PCT/CN2022/120921 CN2022120921W WO2023087915A1 WO 2023087915 A1 WO2023087915 A1 WO 2023087915A1 CN 2022120921 W CN2022120921 W CN 2022120921W WO 2023087915 A1 WO2023087915 A1 WO 2023087915A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
application
screen element
touch operation
user interface
Prior art date
Application number
PCT/CN2022/120921
Other languages
English (en)
French (fr)
Inventor
陈科鑫
谭维
张晓帆
Original Assignee
杭州逗酷软件科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 杭州逗酷软件科技有限公司 filed Critical 杭州逗酷软件科技有限公司
Publication of WO2023087915A1 publication Critical patent/WO2023087915A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/953Querying, e.g. by the use of web search engines
    • G06F16/9536Search customisation based on social or collaborative filtering
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/95Retrieval from the web
    • G06F16/953Querying, e.g. by the use of web search engines
    • G06F16/9538Presentation of query results

Definitions

  • the present application relates to the field of computer technology, and in particular to an application recommendation method, device, device and computer-readable storage medium.
  • the interaction mode centered on smart devices is gradually developing into a service interaction mode centered on people.
  • the convenience of user operations is improved.
  • application recommendation results are generated by collecting historical behavior data of users on smart devices and combining popular service usage data.
  • the embodiment of the present application expects to provide an application recommendation method, device, device, and computer-readable storage medium, by combining the dual-mode information of screen element information and touch operation information, and considering the interaction data of the user interface from two dimensions , and then combined with the degree of association with multiple applications, the accuracy of application recommendation results is improved.
  • an embodiment of the present application provides an application recommendation method, the method comprising: acquiring touch operation information acting on a user interface; based on the touch operation information, acquiring screen element information corresponding to the user interface; Based on the touch operation information and the screen element information, obtain the degree of association between the screen element information and multiple applications, and the multiple applications are applications of the same type; Recommended display.
  • the embodiment of the present application provides an application recommendation method, the method comprising: acquiring touch operation information for a user interface, and acquiring screen element information corresponding to the user interface; based on the screen element information and the The above touch operation information is jointly learned, and associated with the preset sample fusion information and the preset sample sequence behavior information of each application to obtain the degree of association between the screen element information and multiple applications; Apps that meet the preset conditions to a certain extent are recommended for display.
  • an embodiment of the present application provides an application recommendation device, the device comprising: a first acquisition part configured to acquire touch operation information acting on a user interface; and based on the touch operation information, acquire the The screen element information corresponding to the user interface; the first association part is configured to obtain the degree of association between the screen element information and multiple applications based on the touch operation information and the screen element information, and the multiple applications They are applications of the same type; the first display part is configured to display recommendations for applications whose degree of association meets a preset condition.
  • an embodiment of the present application provides an application recommendation device, the device comprising: a second acquiring part configured to acquire touch operation information for a user interface, and acquire screen element information corresponding to the user interface;
  • the second association part is configured to perform joint learning based on the screen element information and the touch operation information, and perform association processing with the preset sample fusion information and the preset sample sequence behavior information of each application to obtain the The degree of association between the screen element information and multiple applications;
  • the second display part is configured to recommend and display the applications whose degree of association satisfies a preset condition.
  • the embodiment of the present application provides an application recommendation device, the application recommendation device includes a memory and a processor; the memory stores a computer program that can run on the processor, and the processor executes the In the case of a computer program, the application recommendation method described in the first aspect or the second aspect is implemented.
  • the embodiments of the present application provide a computer-readable storage medium, on which executable instructions are stored, configured to implement the application recommendation method described in the first aspect or the second aspect when executed by a processor.
  • Embodiments of the present application provide an application recommendation method, device, device, and computer-readable storage medium.
  • the touch operation information acting on the user interface is obtained; based on the touch operation information, the screen element information corresponding to the user interface is obtained; both the screen element information and the touch operation information are related to the user's current operation
  • the interactive data can meet the user's instant service needs.
  • the degree of correlation between the screen element information and multiple applications is obtained, and the multiple applications are of the same type; and the applications whose correlation degrees meet the preset conditions are recommended and displayed.
  • the embodiment of the present application considers the interaction data of the user interface from two dimensions by combining the dual mode information of the screen element information and the touch operation information, and then combines the degree of association with multiple applications to improve the accuracy of the application recommendation results. sex.
  • FIG. 1 is an exemplary schematic diagram of an application recommendation result provided by an embodiment of the present application
  • FIG. 2 is a flow chart of optional steps of an application recommendation method provided in an embodiment of the present application
  • FIG. 3 is a flow chart of optional steps of another application recommendation method provided in the embodiment of the present application.
  • FIG. 4 is a flow chart of optional steps of another application recommendation method provided in the embodiment of the present application.
  • FIG. 5 is an optional flow chart for calculating hidden vectors of screen elements provided by an embodiment of the present application.
  • FIG. 6 is an optional flow chart for calculating hidden vectors of touch actions provided by an embodiment of the present application.
  • FIG. 7 is a flow chart of optional steps of another application recommendation method provided in the embodiment of the present application.
  • FIG. 8 is an optional flow chart for generating recommendation results provided by an embodiment of the present application.
  • FIG. 9 is a schematic diagram of an exemplary presentation form of an application recommendation result provided by an embodiment of the present application.
  • FIG. 10 is a schematic diagram of an exemplary presentation form of another application recommendation result provided by the embodiment of the present application.
  • FIG. 11 is a flow chart of optional steps of another application recommendation method provided in the embodiment of the present application.
  • FIG. 12 is a schematic structural diagram of an application recommendation device provided in an embodiment of the present application.
  • FIG. 13 is a schematic structural diagram of another application recommendation device provided in the embodiment of the present application.
  • FIG. 14 is a schematic diagram of the composition and structure of an application recommendation device provided in an embodiment of the present application.
  • smart application recommendation devices generate application recommendation results by collecting historical behavior data of users on smart devices and combining popular service data, as shown in Figure 1, which is an application recommendation provided in the embodiment of this application Exemplary schematic diagram of the results.
  • Figure 1 takes the mobile phone as an example to show the application recommendation scene on the mobile phone screen.
  • the content displayed in the "App Suggestion” and “Popular Application” columns in Figure 1 belongs to the application obtained by the intelligent application recommendation device of the terminal through calculation and analysis Recommended results.
  • application suggestion is an application recommendation generated based on the collected user's historical behavior data of smart devices, that is, a service recommendation based on user's historical behavior
  • hot application is generated based on the popular service data connected to the smart device network
  • Application recommendation that is, service recommendation based on popular services, in Figure 1, different letters represent different recommendation contents.
  • the intelligent application recommendation method includes the following characteristics: in the use of recommendation data, the user's historical browsing, click and other interactive data, as well as popular Internet data, are usually used as references when recommending applications to users.
  • the recommendation service In terms of the content of the recommendation service, it mainly provides users with recommendations such as application recommendations, advertisement recommendations, or information messages.
  • the nature of the recommendation service it is mainly to improve the click-through rate and activity rate of the application software. Therefore, it focuses on guiding the user to use the intelligent application to recommend the application software that the device wants the user to use.
  • intelligent application recommendation has at least one or more of the following disadvantages in terms of user service requirements and product performance: (1) Due to the dependence on user historical behavior data and popular service data in related technologies, it is difficult for users It is difficult to cover the immediate service needs of users by recommending the services to be used soon. (2) Since the relevant technology only uses the user's historical behavior data, and does not consider the user's active intention to operate, that is to say, there is no interaction between the intelligent application recommendation device and the user's active behavior, and the user can only receive the application passively. recommend. (3) Since only historical behavior data of users are used in related technologies, the information mode is single, which reduces the accuracy of application recommendation results in related technologies.
  • the embodiment of the present application provides a method for recommending an application, as shown in Figure 2, which is a flow chart of the steps of a method for recommending an application provided in the embodiment of the present application, the method for recommending an application includes the following steps:
  • the user when the user interacts with the application recommendation device, the user operates on the information of interest to the user on the user interface, and touch operation information is generated during the operation.
  • the touch operation information is information applied to the user interface, and the application recommending device can obtain the touch operation information.
  • the user interface is a page when the user interacts with the application recommendation device, and the user interface may be understood as a current page or a current user interface.
  • the touch operation information represents the user's interaction data with respect to the user interface.
  • the touch operation information may include: operation content and touch operation. Touch operations include, but are not limited to, click, double-click, long-press, two-finger pinch, two-finger method, slide, and drag; operations include, but are not limited to, copy, forward, share, drag, remind, save, bookmark, and stay. Among them, staying means that the user operates the controls on the user interface, and the operation content is all content except other user operations.
  • the optional controls in the submenu popped up by the operating system include copy, forward, share, drag and drop, Reminder, save and favorite are taken as examples to illustrate that the user does not select an optional control in the submenu, but selects other controls, or the user selects a control other than the submenu.
  • the touch operation information includes: touch operation and operation content; the above S201 can be implemented through the following three examples.
  • the first example is to obtain the touch operation on the user interface; display the operation content corresponding to the touch operation.
  • the touch operation information when the touch operation information is acquired, it is triggered by the user's touch operation, the touch operation acting on the user interface is acquired first, and the operation content corresponding to the touch operation is displayed in the user interface.
  • the touch operation is a touch gesture as an example for illustration.
  • the user needs to perform a copy operation on certain text information.
  • the user's touch gesture is a "long press”
  • an Operation content the operation content is copy, share, forward, save, favorite, etc., and the user selects the operation content "copy”.
  • the touch operation acting on the user interface is obtained, and the operation content corresponding to the touch operation is displayed. If the user selects the operation content, the operation content is obtained, thereby obtaining the touch operation information. Compared with directly obtaining The scheme of touch operation information reduces resource consumption.
  • the touch operation information includes: touch operation and operation content; the above S201 is implemented through the following second example.
  • the touch operation acting on the user interface is obtained; based on the touch operation, the screen element information corresponding to the user interface is obtained; based on the touch operation and the screen element information, the corresponding operation content is displayed.
  • the touch operation information when the touch operation information is acquired, it is triggered by the user's touch operation, and the touch operation reflects that the user needs to operate some content in the user interface in order to select the operation content next, and the screen element information Reflect the supported operation content. Therefore, the displayed operation content is content related to the touch operation and screen element information, and the corresponding operation content is displayed based on the touch operation and screen element information, thereby improving the accuracy of the operation content.
  • the touch operation is a touch gesture as an example. The user needs to perform a copy operation on a certain text information. When the user's touch gesture "long press" is detected, the screen element information corresponding to the user interface is obtained. The operation content is generated based on the position of the touch gesture and the screen element information. The operation content is copy, share, forward, save, favorite, etc., and the user selects the operation content "copy”.
  • the touch operation acting on the user interface is obtained first, and based on the touch operation, the screen element information corresponding to the user interface is obtained; based on the touch operation and screen element information, the corresponding operation content is displayed. If selected, the operation content is obtained, so as to obtain the touch operation information. Compared with the solution of directly obtaining the touch operation information, resource consumption is reduced.
  • the touch operation information includes: touch operation and operation content; the above S201 is implemented through the following third example.
  • the operation content corresponding to the selected content is displayed; within a preset time range, a touch operation on the operation content is acquired; the touch operation includes the selection operation.
  • the operating system when the user interacts with the application recommendation device, the operating system will always generate data, including touch operation information and screen element information.
  • the embodiment of the application does not always obtain touch operation information and screen element information. information, and will not always recommend applications to users, and the operating system will not always obtain touch operations and screen element information. It can be understood that the user can always perform touch operations on the application recommendation device, but the application recommendation device does not respond to all touch operations.
  • the application recommendation method in the embodiment of the present application is an application recommendation triggered by operation content.
  • the operation content corresponding to the selected content is displayed; thus, within the preset time range, the touch operation on the operation content is acquired, compared with always acquiring touch
  • the method of operation reduces resource consumption.
  • the information of the screen elements corresponding to the user interface changes.
  • the operating system will always generate data, including touch operation information and screen element information, and the application recommendation device does not always recommend applications to users.
  • the screen element information corresponding to the user interface is obtained, and the screen element information is a screen element corresponding to the user touch operation information, thereby improving the accuracy of the screen element information.
  • the screen element information represents information of all elements displayed in the user interface, including but not limited to control information, text information, and image information.
  • structural knowledge represents the connection between entities, and can also be understood as the connection between controls
  • textual knowledge represents text content such as movies, text or links
  • visual knowledge represents Image content such as covers or posters.
  • determining the current application service set by performing service analysis on the current user interface can be understood as selecting applications related to the current user interface as the current application service set, reducing the number of applications that need to be coded, thereby reducing resource consumption.
  • the area to be identified may be selected from all areas corresponding to the user interface by those skilled in the art according to actual needs, or the area to be identified may be determined according to the user's operation area, thereby improving the accuracy of the area to be identified. The accuracy is not limited by this embodiment of the present application.
  • the above-mentioned control information identification is required when obtaining the screen element information corresponding to the user interface.
  • the user interface does not necessarily include textual knowledge, visual knowledge, and the current application service set. Therefore, textual knowledge, visual knowledge, and current application service At least one of the set and the structural knowledge are used as the screen element information, which improves the richness and accuracy of the screen element information.
  • the following steps may be included: Obtain the structural information of the control tree corresponding to the user interface or the area to be identified through the accessibility service interface ; Identify the control information on the structural information of the control tree to obtain structural knowledge.
  • the accessibility service is a set of system-level application programming interfaces (Application Programming Interface, API) that can simulate operations. After the user agrees to the app's permission to obtain the accessibility service, the user can simulate the operation and control the user's app recommendation device in turn.
  • the barrier-free service can be used in applications such as grabbing red envelopes, automatic replies, and one-click access to permissions to achieve one-click operations.
  • the identification method of the control information can use the barrier-free service interface provided by the operating system to obtain the structural information of the entire control tree corresponding to the user interface or the area to be identified.
  • the user interface or the area to be identified When the user interface or the area to be identified is refreshed, it can be called without Obstacle service interface, to obtain the structure of the entire control tree of the updated interface. Then, according to the structural information of the entire control tree, the control information is identified for the structural information of the control tree through topology technology, and the structural knowledge is obtained.
  • the structural information of the control tree corresponding to the user interface or the area to be identified is obtained through the barrier-free service interface; the control information is identified for the structural information of the control tree to obtain structural knowledge, which improves the accuracy of structural knowledge. accuracy.
  • the application recommendation device is to meet the user's immediate service needs, so it is necessary to obtain touch operation information of the user interface and screen element information corresponding to the user interface.
  • Screen element information and touch action information reflect strong intentions related to the user's immediate needs, which can be understood as user-related instant information. After obtaining the instant information, it is necessary to calculate the degree of correlation between the screen element information and multiple applications.
  • multiple applications may be applications of the same type.
  • the same type of application can be understood as the same type of expression of the functions corresponding to the application, for example, multiple different playback software used to play images or videos to users are applications of the same type; multiple different playback software used for shopping The shopping software is the same type of application; multiple different social networking software used for social networking are the same type of application.
  • multiple applications can be determined based on the touch operation information and screen element information, and the multiple applications are applications of the same type, and all applications of this type can support the operation content and screen elements in the touch operation information in the current user interface Information, and then perform collaborative joint learning and similarity processing between the screen element information and multiple applications to obtain multiple degrees of association.
  • collaborative joint learning is a process of fusing hidden information of screen elements and touch action hidden information, and similarity processing is to combine the fused information with collaborative filtering.
  • the collaborative filtering algorithm is based on mining the user's historical behavior data to discover the user's preference bias, and predict the products that the user may like to recommend. It can be understood as functions such as "guess you like it” and "people who bought this product also like it”. It can be realized in the following ways: recommending you based on people who share your preferences, recommending similar items based on your favorite items, and comprehensively recommending based on the above conditions.
  • an application includes: an application type, and/or, an application service.
  • the application can be an application type, an application service, or a combination of an application type and an application service.
  • the application type can be understood as the type of application software, such as the application listed in Figure 1 Recommendations and individual app services for popular apps.
  • Application services can be understood as displaying text or images in the user interface in the application software, for example, opening images, videos and other files in the playback software, opening the shopping link in the shopping software, opening ordinary links in the browser, Make calls to other users in dial-up software, and open documents in office software.
  • the application recommendation result may also be a combination of application type and application service.
  • the application recommendation in the embodiment of the present application may also be understood as service recommendation, application service recommendation, application software recommendation, etc., which is not limited in the embodiment of the present application.
  • the application or the application recommendation result includes the application type and/or application service, which increases the richness of the application recommendation result.
  • the touch operation information and screen element information are related to the current user interface, which can meet the immediate needs of users. Based on the touch operation information and screen element information, multiple applications of the same type can be determined. The degree of association between screen element information and all applications is obtained, and the degree of association between screen element information and multiple applications of the same type is acquired, thereby reducing resource consumption and improving the accuracy of the degree of association.
  • the application corresponding to the degree of association that satisfies the preset condition is used as the application recommendation result, and the application is recommended and displayed.
  • a sign indicating whether to open the application is firstly displayed to the user.
  • the sign may be displayed in the form of a window, text, picture, folder, etc. Taking a window as an example, if the user clicks on the window, the displayed association degree satisfies the preset conditional application. It is also possible to directly display to the user the applications whose degree of relevance satisfies the preset condition, and the user may directly select the applications.
  • the preset condition can be appropriately set by those skilled in the art according to actual needs. For example, if the preset condition is that the degree of relevance is greater than the preset threshold, one or more applications will be recommended to the user; if the preset condition is the maximum correlation degree, an application is recommended to the user, which is not limited in this embodiment of the present application.
  • the above S204 may be implemented in the following manner. On the user interface, it is recommended to display the application with the highest similarity among the applications; or, on the user interface, it is recommended to display the preset number of applications with the highest similarity among the applications.
  • the application with the highest display similarity is recommended to the user interface.
  • the application is directly entered, and the user can jump to the application without further operation, which improves the efficiency of application recommendation.
  • the application is recommended to the user on the user interface, and the user only needs to choose whether to open or enter the application, and the user does not need to select among multiple applications, which improves the efficiency of application recommendation.
  • the preset number of applications with the highest similarity among the applications are recommended, and the preset number of applications are recommended to the user.
  • the user can select applications that are strongly related to their own needs, which improves the diversity of application recommendations.
  • the preset number can be appropriately set by those skilled in the art according to the actual situation, for example, 2, 3, 4, etc., which is not limited in this embodiment of the present application.
  • the touch operation information acting on the user interface is obtained; based on the touch operation information, the screen element information corresponding to the user interface is obtained; both the screen element information and the touch operation information are related to the user's current operation
  • the interactive data can meet the user's instant service needs.
  • the degree of correlation between the screen element information and multiple applications is obtained, and the multiple applications are of the same type; and the applications whose correlation degrees meet the preset conditions are recommended and displayed.
  • the embodiment of the present application considers the interaction data of the user interface from two dimensions by combining the dual mode information of the screen element information and the touch operation information, and then combines the degree of association with multiple applications to improve the accuracy of the application recommendation results. sex.
  • the touch operation information includes operation content; acquiring screen element information corresponding to the user interface; the above S202 can be implemented through the following two examples.
  • the touch operation information is satisfied, which indicates that the operation content satisfies the preset trigger recommendation condition, the screen element information of the entire interface corresponding to the user interface is obtained; the preset trigger recommendation condition indicates the expected operation intention.
  • the application recommendation device does not always recommend applications to the user. If the touch operation information indicates that the operation content does not meet the expected operation intention, it means that the user has no strong intention. At this time, no application recommendation will be made to the user, and there is no need to obtain the screen element information of the entire interface corresponding to the user interface. . When the touch operation information is satisfied to indicate that the operation content has reached the expected operation intention, indicating that the user has a strong intention to super share, forward or open, etc., then obtain the screen element information of the entire interface corresponding to the user interface.
  • the full interface corresponding to the user interface can be understood as the entire area of the user interface, and the preset trigger recommendation conditions can be appropriately set by those skilled in the art according to actual needs, and it only needs to be able to distinguish whether the user has strong intentions.
  • the preset triggering recommendation condition represents an expected operation intention.
  • the preset triggering condition may be a preset operation content
  • the preset operation content may be copying, forwarding, sharing, dragging, reminding, Save or bookmark
  • the preset trigger recommendation condition may be at least one of the following: copy, forward, share, drag, remind, save and bookmark, but this embodiment of the present application is not limited.
  • the preset trigger condition is "copy” in the operation content
  • the user's touch operation on a link is "long press”
  • the operating system pops up a submenu
  • the user operation content is to select the link in the submenu
  • “Copy” means that the user has a strong intention to super share, forward or open, etc., indicating that the user's operation content meets the preset trigger recommendation conditions.
  • the above S202 is implemented through the following second example.
  • the current area to be identified is determined at the touch position where the touch operation information acts; the current area to be identified is identified to determine screen element information.
  • the screen element information is information for different types of elements displayed in the user interface
  • the touch position of the touch operation information is part of the interface in the user interface, that is to say, the user touch operation is for Therefore, according to the touch position of the touch operation information, the current area to be identified is determined.
  • the area of the current area to be identified is smaller than the entire area of the user interface, and then the current area to be identified is identified. Determine the screen element information in the area to be recognized.
  • the current area to be identified is determined according to the touch position of the touch operation information, and then the current area to be identified is identified to determine the screen element information. It is not necessary to obtain the screen element information of all areas of the user interface. The amount of data processing is reduced, and the efficiency of application recommendation is improved.
  • the application recommendation method further includes the following steps: no screen is recognized in the region that satisfies the current region to be recognized In the case of element information, determine the next area to be identified, and perform identification in the next area to be identified until the existence of screen element information or a complete interface is recognized; the next area to be identified is larger than the current area to be identified.
  • the touch position affected by the touch operation information is taken as the center, and the area within the preset range is determined as the current area to be identified. If the screen element information is not recognized in the current area to be identified , indicating that it is necessary to further expand the area of the area to be identified, for example, to expand the preset range, or, based on the current area to be identified, to expand in a certain direction or around, so as to determine the next area to be identified, the next area to be identified greater than the current area to be identified. Then identify the next area to be identified until the information of the existing screen elements is obtained or the interface is completely identified.
  • the area to be identified needs to be expanded until the screen element information is obtained, and the area to be identified is determined in a step-by-step manner, which improves the accuracy of the area to be identified , when the screen element information is obtained, the area to be identified is no longer expanded, and the method of directly identifying all areas of the user interface and obtaining screen element information reduces resource consumption.
  • the touch operation information includes: touch operation and operation content; the above S203 may be implemented in the following manner.
  • the degree of association between the screen element information corresponding to the user interface and the multiple applications is obtained.
  • the application recommendation device displays the corresponding operation content to the user
  • the user selects the operation content to obtain the operation content.
  • the degree of association between the screen element information corresponding to the user interface and the multiple applications is obtained.
  • the application recommendation method is triggered by the operation content.
  • multiple applications are determined according to the operation content, touch operation, and screen element information. Multiple applications can support touch in the current user interface. Control the application of operation information and screen element information, so as to obtain the correlation degree between screen element information and multiple applications. Compared with directly obtaining the correlation degree between screen element information and all applications, it reduces resource consumption and improves the accuracy of correlation degree. .
  • the above S203 may also be implemented in the following manner. Based on the screen element information and touch operation information, correlate with the preset sample fusion information and the preset sample sequence behavior information of each application on each application, so as to obtain the degree of correlation between the screen element information and each application of multiple applications .
  • the preset sample fusion information and the preset sample sequence behavior information of each application can be understood as the preset training data, which is the user and mobile phone information obtained before the application recommendation.
  • the data generated by the interaction can be used as a reference for this application recommendation.
  • the preset training data includes a plurality of training samples, and after knowledge base embedding and collaborative joint learning are performed on the preset training data, preset sample fusion information is obtained.
  • the preset training data also includes sample service information of multiple training samples. There will be the same application in multiple sample service information, and the preference probability of each application is predicted according to the multiple sample service information, so as to obtain the preset Sample sequence behavior information.
  • a training sample includes a piece of sample fusion information and sample service information corresponding to the sample fusion information.
  • the sample service information represents the user's subsequent selection of each piece of interaction data (the interaction data corresponds to the sample fusion information) in the training sample.
  • the sample service information can be Understand as applying the truth value.
  • the sample sequence behavior information of each application is obtained according to multiple pieces of sample fusion information and the sample service information corresponding to each piece of sample fusion information.
  • the sample sequence behavior information of each application represents the user's preference probability for each application for different sample fusion information.
  • the sample sequence behavior information of the i-th sample fusion information for the j-th application can also be understood as the preference probability of the i-th sample fusion information for the j-th application.
  • the preference probability is determined by the number of sample service information corresponding to the sample fusion information in the training data. It can also be understood as the user's preference probability for each application under the same sample fusion information. The higher the preference probability, the greater the value of the sample sequence behavior information .
  • the i-th sample fusion information will affect the first application
  • knowledge base embedding and collaborative joint learning are performed based on screen element information and touch operation information to obtain fusion data to be analyzed, and then merge information with preset samples on each application and each preset application
  • the behavior information of the sample sequence of the sample sequence is used for collaborative joint learning and correlation processing, so as to obtain the correlation degree between the screen element information and each application of multiple applications, and improve the accuracy of the correlation degree.
  • the application recommendation method further includes the following steps: according to the preset training data, determine the sample screen element hidden information, sample touch action hidden information and sample service information corresponding to each training sample; Perform collaborative joint learning on the hidden information of sample screen elements and sample touch action hidden information to obtain the preset sample fusion information of each training sample; determine the preset sample sequence behavior information of each application according to the sample service information of each training sample .
  • the preset training data includes a plurality of training samples
  • one training sample includes sample screen element information, sample touch action information, and sample service information.
  • the hidden information of the sample screen element and the sample touch action hidden information of each training sample are respectively analyzed and encoded by the knowledge base embedding method, Then through collaborative joint learning, the hidden information of the sample screen elements and the hidden information of the sample touch actions are fused to obtain the sample fusion information.
  • the number of times (or probability) that each application is selected is obtained, thereby determining the sample sequence behavior information of each application .
  • the sample fusion information of each training sample and the sample sequence behavior information of each application are determined, so as to carry out collaborative joint learning with the screen element information and touch operation information corresponding to the user interface and similarity processing to obtain the relevance of each application.
  • each application is associated with the preset sample fusion information and the preset sample sequence behavior information of each application, so as to obtain the screen element information and multiple
  • the degree of relevance of each application of the application may include S301-S303.
  • FIG. 3 is a flow chart of optional steps of another application recommendation method provided in an embodiment of the present application.
  • an auto encoder can be used to encode the screen element information to obtain the hidden information of the screen element.
  • the auto encoder is an artificial neural network capable of learning efficient representation of input data through unsupervised learning. Networks, this efficient representation of input data is called codings, and its dimensions are generally much smaller than the input data, making autoencoders useful for dimensionality reduction.
  • the autoencoder can be any form of encoder, and there is no restriction on the structure of the autoencoder used, as long as it can encode the screen element information and touch operation information separately, including but not limited to vanilla Autoencoders, Multilayer Autoencoders, Convolutional Autoencoders, and Regularized Autoencoders.
  • regular autoencoder includes sparse autoencoder and denoising autoencoder
  • denoising autoencoder can be stacked denoising autoencoder (stacked denoising auto-encoders, SDAE)
  • convolutional autoencoder can be stacked volume Product self-encoder (stacked convolutional auto-encoders, SCAE).
  • an autoencoder usually consists of two parts: the encoder (also known as the recognition network) converts the input into an internal representation, the decoder (also known as the generation network) converts the internal representation into an output, and the output is trying to reconstruct the input.
  • the loss function is reconstruction loss.
  • the encoding part of the autoencoder i.e. encoder
  • the decoding part of the autoencoder i.e. decoder
  • the image to be encoded is encoded by using the encoding part of the trained autoencoder (ie encoder) to obtain a high-dimensional vector, which may include all information of the encoded image.
  • the touch operation information may be encoded in a common encoding manner to obtain the touch action hidden information.
  • the embodiment of the present application does not limit the coding method, as long as the coding method can distinguish which touch operations are performed by the user.
  • a 0-1 encoding method is used for encoding, 1 indicates that the current user has performed the touch operation, and 0 indicates that the current user has not performed the touch operation.
  • other forms of encoding methods may also be used to distinguish between the touch operation performed and the non-conducted touch operation, which is not limited in the embodiment of the present application.
  • the screen element information is the data related to the user interface
  • the touch operation information is the interaction data related to the user's current operation
  • different encoding forms are used to encode the dual-modal information (screen element hidden information and touch action hidden information)
  • the obtained screen element hidden information and touch action hidden information are interactive data related to the user's current operation, which can meet the user's immediate service needs.
  • the correlation of each application is obtained, and the accuracy of each correlation is improved.
  • the screen element information includes: structural knowledge corresponding to the user interface; the above S301 may be implemented in the following manner. Encode the structural knowledge to obtain the hidden information of screen elements.
  • the embodiment of the present application encodes the structural knowledge to obtain structural information, and uses the structural information as hidden information of screen elements.
  • the TransR model can represent structural type characteristics.
  • the structural knowledge is encoded to obtain the hidden information of the screen elements, which improves the accuracy of the hidden information of the screen elements.
  • the screen element information includes: the structural knowledge, textual knowledge, visual knowledge corresponding to the user interface, and the current application service set related to the user interface; the current application service set is determined by analyzing the service of the user interface .
  • the above S301 in FIG. 3 may include S3011-S3013, as shown in FIG. 4 , which is a flow chart of optional steps of another application recommendation method provided in an embodiment of the present application.
  • S3011. Encode the structural knowledge, textual knowledge and visual knowledge respectively to obtain structural information, textual information and visual information.
  • the hidden information of screen elements is determined by data embedding method.
  • the data embedding method may be a collaborative knowledge base embedding (Collaborative Knowledge base Embedding, CKE) algorithm, wherein, the CKE algorithm includes knowledge base embedding (knowledge base embedding) and collaborative joint learning, wherein, the knowledge base embedding can also be understood as Knowledge graph embedding.
  • CKE collaborative Knowledge base Embedding
  • the knowledge base embedding method is used to determine the hidden information of screen elements.
  • collaborative joint learning can be used to fuse the hidden information of screen elements and hidden information of touch actions to obtain the fusion information to be analyzed.
  • the encoding process of visual information, text information and structural information will be described respectively below.
  • the structural knowledge is encoded to obtain the structural information, which is specifically described above and will not be repeated here.
  • the embedding method that can be used is the stacked convolutional autoencoder SCAE.
  • SCAE uses stacked convolutions during training.
  • the neural network maps (encodes) the information of the original image to a high-dimensional vector, and then uses the high-dimensional vector to reconstruct (decode) the original image. Therefore, the high-dimensional vector includes all the information of the original image.
  • the coding part of the trained SCAE is used to code the visual knowledge to obtain a high-dimensional vector.
  • the high-dimensional vector is concatenated with the encoded vector of the image source analysis result to obtain the visual vector.
  • the encoded text information as an expression of text vectors as an example.
  • the embedding method that can be used is the stacked denoising autoencoder SDAE.
  • the idea of SDAE is similar to that of SCAE. Let me repeat.
  • the text-type knowledge is encoded by using the encoding part of the trained SDAE to obtain a high-dimensional vector, and the high-dimensional vector is spliced with the encoding vector of the additional information of the text to obtain a text vector.
  • an autoencoder of any structure can be used to encode structural knowledge and textual knowledge.
  • the above examples are only illustrations using SDAE and SCAE as examples, and do not mean that the embodiment of the present application is limited thereto.
  • the structural knowledge in S3011 includes: at least one of control type, control function, and control structure distribution; textual knowledge includes: at least one of character recognition information and special symbol recognition information, and text Original information; visual knowledge includes: at least one of image content identification information and image source information, and original image information.
  • control information is identified to obtain the control type, control function, control structure distribution, etc. on the user interface.
  • types of controls include buttons, texts, labels, and image points, etc.
  • control structure distribution includes the distribution of selection buttons.
  • character recognition information, link recognition information, and special symbol recognition information are obtained from text content recognition, and, for texts containing additional information, for example, a shared link from an application software will be added Information is also encoded into the recognition results, for example, the extra information corresponding to normal URLs and shopping links is different. Therefore, compared with structural knowledge, textual knowledge also includes the original information of the text, which can be understood as the information of the text itself, which can not only reflect characters and special symbols, but also reflect additional information such as links.
  • image content identification is the analysis and description of image content (for example, the description of the content of the image itself), and image content labeling (for example, identifying which specific tourist attraction the landscape photo in the image is, whether the image is emoticons, etc.).
  • Image content recognition may also include image source analysis, which analyzes the source of the image according to the style data of the image (for example, analyzing whether the image comes from a dial-up interface, analyzing whether the image is a screenshot from a certain application software interface). Therefore, compared with structural knowledge, visual knowledge also includes the original information of the image, which can be understood as the information of the image itself, which can not only reflect the image content and image label, but also reflect the original information such as the source of the image.
  • structural knowledge only includes the recognized structural data, that is to say, structural knowledge does not include the substantive content related to user interaction in the user interface, while textual knowledge includes both the recognized character recognition information and Special symbol recognition information also includes original text information, and visual knowledge includes not only recognized image content identification information and image source information, but also original image information.
  • structural knowledge includes at least one of control type, control function, and control structure distribution
  • textual knowledge includes at least one of character recognition information and special symbol recognition information, as well as original text information
  • visual knowledge Knowledge includes at least one of image content identification information and image source information, as well as original image information, which improves the richness of structural knowledge, textual knowledge, and visual knowledge.
  • S3011 in FIG. 4 may include the following steps: perform vectorized coding on structural knowledge to obtain structural information; perform vectorized coding on at least one of character recognition information and special symbol recognition information to obtain text content Comprehend the coding information; and stack self-encoding on the original information of the text to obtain high-dimensional information of the text; splicing the high-dimensional information of the text and the coding information of the text content to obtain the text information;
  • One is to perform informatization encoding to obtain image content understanding encoding information; and to perform stacked self-encoding on original image information to obtain image high-dimensional information; to splicing image high-dimensional information and image content understanding encoding information to obtain visual information.
  • the structural knowledge since the structural knowledge includes at least one of the control type, control function and control structure distribution, there is no substantive content related to user interaction. Therefore, when encoding the structural knowledge, the structural knowledge
  • the structure information can be obtained by performing vectorized coding, which improves the coding efficiency of the structure information.
  • textual knowledge includes at least one of character recognition information and special symbol recognition information, as well as original text information
  • At least one of the identification information is vectorized and coded to obtain text content understanding coding information; and the original text information is stacked and self-encoded to obtain high-dimensional text information, which can reflect the original text information.
  • the text information is obtained by splicing the high-dimensional information of the text and the understanding and encoding information of the text content, which improves the accuracy of the text information.
  • visual knowledge includes at least one of image content identification information and image source information, as well as original image information
  • image content identification information and image Informatization encoding is performed on at least one of the source information to obtain image content understanding encoding information
  • stacked self-encoding is performed on the original image information to obtain image high-dimensional information
  • the image high-dimensional information can reflect the original image information.
  • the high-dimensional image information and image content understanding and encoding information are spliced to obtain visual information, which improves the accuracy of visual information.
  • the service set of the current application is directly and simply coded to obtain the service bias information.
  • the current user interface is a chat interface of A
  • the chat interface of A for a If the user's touch operation is "long press” and the operation content is "copy”, then the application service collection related to the current A chat interface can include another B application software (used in another social software) link sharing), C application software (used to open the shopping link and learn more about the items in the link), D application software (used to query on the website).
  • B application software used in another social software
  • C application software used to open the shopping link and learn more about the items in the link
  • D application software used to query on the website.
  • determining the current application service set by performing service analysis on the current user interface can be understood as selecting applications related to the current user interface as the current application service set, reducing the number of applications that need to be coded, thereby reducing resources consume. Since the current application service needs to encode data that is fixed and the update frequency is low, for example, a certain browser supports opening and parsing a certain type of link, so the current application service collection is directly and simply encoded to obtain service bias information. Use different numbers to distinguish applications.
  • the application is an application service, and the service bias information is a vector expression form.
  • the application service A is encoded to obtain the service bias vector 0001, and the application Service B performs encoding to obtain service bias vector 0010.
  • the service bias information and the screen element encoding information are spliced together to obtain the hidden information of the screen element.
  • S3013 Concatenate at least one of text information, visual information, service bias information, and structural information to obtain screen element hidden information.
  • some user interfaces do not have images and texts, that is, screen element information does not include text information and visual information; some user interfaces do not have application service sets related to the user interface, that is, the current application service set Is empty.
  • the user interface includes structural information; or, the user interface includes structural information and at least one of text information, visual information, and service bias information. Therefore, when obtaining the hidden information of screen elements, for example, at least one of text information, visual information and service bias information, and structural information may be spliced to obtain hidden information of screen elements. It is also possible to splicing structure information, text information, visual information, and service bias information to obtain screen element hidden information; this embodiment of the present application does not limit it.
  • the structural knowledge, textual knowledge, visual knowledge and current application service set are coded respectively, and the coded information is spliced to obtain the hidden information of screen elements, which improves the screen element accuracy of hidden information.
  • screen element hidden vector acquisition module an exemplary application of the embodiment of the present application in an actual application scenario is described through the screen element hidden vector acquisition module.
  • the expression forms of screen element hidden information, touch action hidden information, service bias information, structural information, visual information, and text information are all vectors, and the screen element information is mobile phone screen element information as an example.
  • the screen element hidden vector acquisition module is used to acquire the vector expression form of the mobile phone screen element of the user interface, and the vector expression form here can be regarded as a vector embedding expression based on knowledge graph.
  • FIG. 5 is an optional flow chart for calculating hidden vectors of screen elements provided by the embodiment of the present application.
  • Figure 5 takes the mobile phone as an example for the application recommendation device.
  • the data set used in calculating the hidden vector of the screen element in Figure 5 includes the screen element information corresponding to the user interface of the current mobile phone screen, and the user interface of the current mobile phone. A collection of application services provided.
  • the mobile phone screen element information is analyzed through the following three dimensions: control information identification, text content identification and image content identification.
  • structural knowledge represents the tree-structured knowledge obtained from control information recognition
  • textual knowledge represents the information recognized by text content
  • visual knowledge represents the information recognized by image content.
  • the analysis and identification of the above mobile phone screen element information is triggered by the user operation content in the user touch action.
  • the above analysis and identification step of the mobile phone screen element information is triggered.
  • control information, image content and text content are all identified and coded. If there is no image and/or text in the user interface, the code vector corresponding to the image content and/or text content is to represent that there is no
  • the content vector can be understood as the encoding vector is empty, for example, the encoding vector is 0000.
  • the visual knowledge in Fig. 5 is transformed into corresponding encoding vectors through visual embedding, textual knowledge through text embedding, and structural knowledge through structural embedding. Includes visual vectors, text vectors, and structural vectors. And the current application service set is simply coded to obtain the service bias vector, for example, different numbers are used to distinguish the applications. Concatenate the service bias vector, visual vector, text vector and structure vector to obtain the hidden vector of the screen element.
  • the touch operation information includes: operation content and touch operation; when the touch operation information is encoded to obtain the touch action hidden information, it may be implemented in the following manner.
  • the touch operation and operation content are respectively coded to obtain touch operation information and operation content information; the touch operation information and operation content information are spliced to obtain touch action hidden information.
  • common coding methods can be used to code the operation content and the touch operation, as long as the operations and gestures performed by the user can be distinguished.
  • touch operations include but are not limited to click, double-click, long-press, two-finger pinch, two-finger zoom-in, slide, and drag; use 0-1 coding method for coding, 1 means the current user has made this gesture, 0 It means that the user has not made the gesture at present. In the embodiment of the present application, other encoding methods may also be used to distinguish whether the gesture has been made. This embodiment of the present application does not limit it.
  • the operation content includes, but is not limited to, copying, forwarding, sharing, dragging, reminding, saving, bookmarking, and staying.
  • Use 0-1 encoding method for encoding 1 means that the current user has selected this operation, 0 means that the current user has not selected this operation, in the embodiment of this application, other forms of encoding methods can also be used to determine whether this item is selected operations are differentiated. This embodiment of the present application does not limit it.
  • the operation content information and the touch operation information are spliced by splicing to obtain the hidden information of the touch action. Improved the accuracy of touch action hidden information.
  • the expression form of the touch action hidden information is a vector
  • the touch operation information is a user touch action as an example for illustration.
  • the touch action hidden vector acquisition module is used to acquire the vector expression form of the user touch action
  • the user touch action includes the user operation content and the user touch operation, as shown in Figure 6, which is the An optional flow chart for calculating hidden vectors of touch actions provided by the embodiment.
  • the touch gestures in Figure 6 represent touch operations.
  • the operation gestures supported by user touch gestures include but are not limited to click, double-click, long press, two-finger pinch, two-finger zoom, slide, and drag; user operations supported by user operation content include But not limited to copying, forwarding, sharing, dragging, reminding, saving, favorite and staying, etc.
  • the user operation content vector and the user touch gesture vector are spliced together by vector splicing to obtain the touch action hidden vector.
  • S303 in FIG. 3 may include S3031-S3033.
  • FIG. 7 is a flow chart of optional steps of another application recommendation method provided in the embodiment of the present application.
  • cooperative joint learning is a process of fusing hidden information of screen elements and hidden information of touch actions to obtain fused information to be analyzed.
  • the main function of collaborative joint learning is feature mapping and feature dimensionality reduction.
  • the cooperative joint learning method is used to realize the hidden information of touch actions and hidden information of screen elements as input, and output the fusion information representing the user's intention, that is, the fusion information to be analyzed.
  • the process of collaborative joint learning can be understood as the process of using the collaborative joint model to realize latent information fusion.
  • the latent information of screen elements and touch action hidden information are input into the trained collaborative joint model, and the fusion information is output.
  • the output fusion Information is the information to be analyzed and fused.
  • the collaborative joint model in the embodiment of this application can be any form of network. There is no restriction on the structure of the collaborative joint model used, as long as it can fuse the hidden information of screen elements and touch action hidden information. , to get the correct representation of the fused information.
  • the mapping structure is a fully convolutional network (FCN).
  • the historical behavior data is the user's historical behavior data, which can reflect the user's preference probability for each application.
  • the historical sequence behavior information of each application is obtained.
  • historical sequence behavior information is generated according to the user's historical behavior data, which is denoted as w k , thus m pieces of historical sequence behavior information can be generated, and the historical sequence behavior information w
  • the dimension of k is 1 ⁇ m.
  • S3032 in FIG. 7 may be implemented in the following manner. Based on the historical behavior data, determine the historical behavior parameters of multiple applications; for each application, set the historical behavior parameters corresponding to the application as preset parameters, and combine the historical behavior parameters of multiple applications except the application to get Historical sequence behavior information of the application.
  • the historical behavior parameters corresponding to the applications are preference probabilities
  • the preset parameters are preset probabilities. If the user is a new user, there is no historical behavior data of the user, and there is no preference record of the user, then the preference probabilities of the m applications in the historical sequence behavior information are equal. If the user is not a new user, calculate the preference probabilities of m applications according to the historical behavior data, that is, calculate the probability that the user chooses each application in the historical behavior data.
  • the preset parameters can be properly set by those skilled in the art according to actual needs, and it is only necessary to effectively calculate the recommendation score for the application, and by analogy, m pieces of historical sequence behavior information can be obtained.
  • the behavior data corresponding to the application is set as a preset parameter, and the historical sequence behavior information of the application is obtained by combining the behavior data except the application in the historical behavior data, which is convenient Subsequently, the historical sequence behavior information of the application is used to calculate the similarity to improve the accuracy of the similarity.
  • similarity processing is performed based on historical sequence behavior information and sample sequence behavior information of each application, and similarity processing is performed based on fusion information to be analyzed and sample fusion information of each training sample. Due to the one-to-one correspondence between the sample fusion information and the sample sequence behavior information in the training samples, according to the above two similarity processing, the similarity of each application can be obtained, which improves the accuracy of the similarity.
  • S3033 in FIG. 7 may include the following steps: performing similarity processing based on the historical sequence behavior information and sample sequence behavior information of each application, and determining the first similarity information of each application for each training sample; to be analyzed performing similarity processing on the fusion information and the sample fusion information of each training sample to obtain second similarity information corresponding to each training sample; The similarity information is used to obtain the similarity of each application.
  • n the number of training samples is n
  • n pieces of first similarity information can be obtained.
  • the similarity between the fusion information to be analyzed and the fusion information of each training sample If the number of training samples is n, then n pieces of second similarity information can be obtained.
  • the kth application similarity is obtained.
  • the fusion information and the sequence behavior information are expressed as vectors and the similarity information is a similarity score as an example for illustration.
  • the similarity score can be represented by Pearson correlation coefficient. As shown in formula (1):
  • the similarity of the vectors x1 and x2 in the above formula (1) is expressed as sim(x1, x2), and the Pearson correlation coefficient is defined as the product of the covariance of two vectors divided by their standard deviations.
  • cov(x1,x2) and E[(X1- ⁇ x1)(X2- ⁇ x2)] represent the covariance of two vectors.
  • X1 and X2 represent two vectors respectively
  • ⁇ x1 and ⁇ x2 represent two vectors respectively
  • ⁇ x1 and ⁇ x2 represent the standard deviation of the two vectors, respectively.
  • the similarity calculation is performed based on historical sequence behavior information and sample fusion information, and the similarity information of two dimensions is comprehensively considered to obtain the similarity of each application, which improves the accuracy of the similarity.
  • the following steps may be included: The first similarity information of each training sample, and the second similarity information corresponding to each training sample, obtain the third similarity of each application for each training sample; for each application, sum the third similarity of each training sample , to get the similarity of each application.
  • the first similarity information of the training samples is multiplied by the second similarity information corresponding to the training samples to obtain the third similarity, because the first similarity
  • the numbers after the degree information and the second similarity information are both n, therefore, n third similarities can be obtained after multiplication, one training sample corresponds to one third similarity, and the n third similarities are summed, Get the similarity of each application.
  • the expression form of the fusion information is a vector
  • the similarity information is a similarity score
  • the historical sequence behavior information and the sample sequence behavior information are both operation sequence vectors
  • the similarity is a recommendation score
  • the training samples correspond to each
  • the sample sequence behavior information of the application is a sequence behavior matrix as an example for illustration.
  • the recommendation score of the data to be analyzed for the k-th application service in the application service set can be calculated by the formula (2) Calculation.
  • score in the above formula (2) represents the recommendation score
  • sim(w k , v i ) represents the operation sequence vector w k of the k-th application service
  • sim(s,t i ) represents the similarity score between the fusion vector s to be analyzed and the i-th sample fusion vector ti in the fusion vector set
  • represents the sum of n similarity products.
  • the similarity of each application is obtained, which improves the accuracy of the similarity.
  • a collaborative joint learning module and a recommendation result generating module an exemplary application of the embodiment of the present application in an actual application scenario is described through a collaborative joint learning module and a recommendation result generating module.
  • the expression forms of screen element hidden information, touch action hidden information and fusion information are all vectors, and the sample sequence behavior information corresponding to each application of the training sample is a sequence behavior matrix, and the historical sequence behavior information and sample sequence behavior information are all operation sequences.
  • Vector and similarity information are similarity scores, similarities are recommendation scores, and the relevance of each application is application recommendation results.
  • the process of generating application recommendation results may include: determining The fusion vector representing the user's intention is used to determine the recommendation result with the highest score through the collaborative filtering algorithm.
  • the collaborative joint learning module is used to determine the fusion vector to be analyzed according to the touch action hidden vector and the screen element hidden vector.
  • the latent vector of the touch action and the latent vector of the screen element are used as input, and the fusion vector representing the user's intention is output, that is, the fusion vector to be analyzed.
  • the collaborative joint learning module can be trained according to the inverse gradient of the recommendation results.
  • the collaborative joint learning module in the embodiment of the present application is a trained model, and both the data to be analyzed and the training data need to undergo collaborative joint learning to obtain fusion vectors. The difference is that the data to be analyzed undergoes collaborative joint learning The final result is the fusion vector to be analyzed, and the training sample in the training data is obtained after collaborative joint learning is the sample fusion vector.
  • the application recommendation result generating module is configured to generate an application recommendation result according to the fusion vector to be analyzed and preset training data.
  • Figure 8 is An optional flow chart for generating recommendation results provided by the embodiment of this application.
  • the upper part of Figure 8 is a flow chart of training data, which can represent the data generated by the interaction between the user and the mobile phone, and the lower part of Figure 8 is a flow chart of the data to be analyzed.
  • the application is an application service as an example for illustration.
  • the latent vectors of the sample screen elements and the latent vectors of the sample touch actions can be fused through collaborative joint learning to obtain a sample fusion vector representing the user's intention.
  • the training data includes sample screen element latent vectors, sample touch action latent vectors and sample service information, and the sample fusion vector corresponds to the sample service information one by one.
  • the training data includes n pieces of user interaction data, that is, n training samples, then after the knowledge base embedding and cooperative joint learning are performed, n sample fusion vectors will be obtained.
  • the fusion vector set includes n A sample fusion vector for bar interaction data.
  • the number of sample service information corresponding to n pieces of interaction data is n, and the same application service may exist in the n sample service information.
  • the set of all application services supported by the application recommendation device is taken as an application service set, and the total number of application services included is m, and each application service in the m application services is different, and m is much smaller than n.
  • the knowledge base embedding method is used to analyze and encode the hidden vector of the screen element and the hidden vector of the touch action, as shown in Figure 8 Only the sample screen element latent vectors and sample touch action latent vectors for each training sample in the training data are shown.
  • the training data includes n training samples, and one training sample includes a sample fusion vector and its corresponding sample service information.
  • the sample sequence behavior information of each application service is determined, that is, the sequence behavior matrix whose dimension is n ⁇ m in the upper right corner in Fig. 8 .
  • Rij in Figure 8 represents the preference probability of the i-th sample fusion vector for the j-th application service, and can also be understood as the interaction score of the i-th interaction data for the j-th application service.
  • the interaction score is determined by the number of sample service information corresponding to the sample fusion vector in the training data. It can also be understood as the user’s preference probability (that is, the selection frequency) for each application service under the same interaction data. The higher the preference probability, the higher the Rij value. big.
  • Performing the above calculation on all the interaction data in the fusion vector set can obtain a sequence behavior matrix with a dimension of n ⁇ m. Taking the row vector of each row in the sequence behavior matrix as an operation sequence vector, n operation sequence vectors can be obtained, and the i-th operation sequence vector is denoted as v i , and v n in Figure 8 represents the n-th operation sequence vector.
  • the touch action information and screen element information are obtained, and the knowledge base embedding method is used to analyze and encode the hidden vector of the screen element and the touch action hidden vector.
  • Control the action-hidden vector for collaborative joint learning to obtain the fusion vector to be analyzed denoted as s.
  • the fusion vector s to be analyzed has the same dimension as the i-th sample fusion vector.
  • an operation sequence vector of the data to be analyzed is generated according to the user's historical behavior data, and recorded is w k , thus m operation sequence vectors can be generated.
  • the dimension of the operation sequence vector w k is consistent with the dimension of the row vector of each row in the above sequence behavior matrix, that is, the dimension of the i-th operation sequence vector v i is consistent, that is, 1 ⁇ m.
  • the combined service module When the combined service module generates the operation sequence vector of the data to be analyzed, for example, for the kth application service among the m application services, the preference probability of the m application services is calculated according to the user's historical behavior data, and the kth application service The preference probability of the application service is set to a preset peak value (that is, a preset parameter), and combined with the remaining m-1 application service preference probabilities, the operation sequence vector w k of the kth application service is determined.
  • m operation sequence vectors can be obtained.
  • the recommendation result generation module can generate the most likely application recommendation by using a collaborative filtering algorithm.
  • a collaborative filtering algorithm For the k-th application service among the m application services, it is necessary to calculate the operation sequence vector w k of the data to be analyzed, and the similarity between the operation sequence vector v i corresponding to each sample fusion vector in the fusion vector set Score, n operation sequence vectors corresponding to n sample fusion vectors, so n similarity scores can be obtained. It is also necessary to calculate the similarity score between the fusion vector s to be analyzed and each sample fusion vector (denoted as t i ) in the fusion vector set.
  • the fusion vector set includes n sample fusion vectors, so n similarity scores can be obtained.
  • the above-mentioned similarity scores of the n operation sequence vectors and the similarity scores of the n fusion vectors are multiplied correspondingly and summed to obtain the recommendation score of the kth application service.
  • the application service with the highest recommendation score is selected to obtain the application recommendation result, and the application recommendation result is obtained by combining the mobile phone screen element information and the user's touch action.
  • the embodiment of the present application proposes an application recommendation method that combines mobile phone screen element information and user touch actions, comprehensively considering the user's specific touch actions and element information corresponding to the screen content that the user is interacting with.
  • the user's strong interaction needs are estimated, and then the corresponding application recommendation is made, which can satisfy the user in real time need.
  • the dual mode information strongly related to the user is used to provide intelligent recommendation services to the user, which improves the accuracy of the application recommendation results.
  • the user's active intention is fully considered in combination with the user's active and strong demand contained in the user's touch action in the intelligent interaction scene and the corresponding screen element information.
  • service which provides an intelligent recommendation service that is actively triggered by user intentions, improving the accuracy of recommendation results. By identifying user intentions and providing personalized and customized services to users, the interaction efficiency between users and devices is improved.
  • S204 in FIG. 2 above may also be implemented in the following manner. Based on the applications meeting the preset conditions, service prompt information is generated; on the user interface, it is recommended to display the service prompt information.
  • service prompt information is generated based on applications that meet preset conditions.
  • the service prompt information may be a list of different types of application software, and the service prompt information is recommended and displayed on the user interface for users to choose.
  • the service prompt information can be displayed at any position in the user interface in the form of a floating window or a button, and the embodiment of the present application does not limit the form and position of the service prompt information.
  • FIG. 9 and FIG. 10 provide a schematic diagram of an exemplary presentation form of an application recommendation result according to an embodiment of the present application.
  • the intelligent service recommendation floating window in Fig. 9 and Fig. 10 is a form of presentation of service prompt information.
  • the current user interface User Interface, UI
  • the application recommendation method of the embodiment of the present application can accurately analyze the potential interaction intention of the user: the user wants to open the sharing link in the shopping application. Therefore, the application recommendation device makes a recommendation service, and displays the application recommendation result in the form of a floating window at the bottom of the mobile phone screen. The user clicks the control 1 of the intelligent service recommendation floating window to jump to the corresponding shopping application A to open the link.
  • the application recommendation device recognizes that the current user interface has received a screenshot sharing from a certain video website provided by Wang Ge's social session, and at the same time detects the user's strong interaction intention: the user has long pressed the link And click to view the original image.
  • the application recommendation method of the embodiment of the present application can identify the potential intention of the user: to open the video in the video application. Therefore, the application recommendation device makes recommendation services, and displays the application recommendation results in the form of a floating window at the bottom of the mobile phone screen. The user clicks the control 2 of the intelligent service recommendation floating window to jump to the corresponding video interface of the video application. video.
  • service prompt information is also generated based on applications that meet the preset conditions; on the user interface, the service prompt information is recommended to be displayed, realizing fast and effective forwarding of the data selected by the user in the user interface.
  • the embodiment of the present application also provides an application recommendation method, as shown in FIG. 11 , which is a flow chart of optional steps of another application recommendation method provided in the embodiment of the present application.
  • the application recommendation method includes the following steps:
  • S1102. Perform joint learning based on screen element information and touch operation information, and perform associated processing with preset sample fusion information and preset sample sequence behavior information of each application to obtain the degree of association between screen element information and multiple applications.
  • the foregoing multiple applications may be applications of the same type, or may be applications of different types, which is not limited in the embodiment of the present application.
  • the above S1102 may further include the following steps: encoding screen element information to obtain screen element hidden information; encoding touch operation information to obtain touch action hidden information; Action hidden information is collaboratively learned to obtain fusion information to be analyzed; historical sequence behavior information of each application determined based on historical behavior data is obtained; similarity processing is performed based on historical sequence behavior information and sample sequence behavior information of each application, and based on The fusion information to be analyzed and the fusion information of each training sample are subjected to similarity processing to obtain the similarity of each application.
  • the above-mentioned similarity processing is performed based on the historical sequence behavior information and sample sequence behavior information of each application, and the similarity processing is performed based on the fusion information to be analyzed and the sample fusion information of each training sample to obtain the similarity of each application.
  • Perform similarity processing based on the historical sequence behavior information and sample sequence behavior information of each application determine the first similarity information of each application for each training sample; perform similarity processing on the fusion information to be analyzed and the fusion information of each training sample, The second similarity information corresponding to each training sample is obtained; based on the first similarity information of each application for each training sample and the second similarity information corresponding to each training sample, the similarity of each application is obtained.
  • the embodiment of the application also provides an application recommendation device, as shown in Figure 12, which is a schematic structural diagram of an application recommendation device provided in the embodiment of the application, the application recommendation The device 120 includes: a first acquiring part 1201 configured to acquire touch operation information acting on the user interface; and based on the touch operation information, acquire screen element information corresponding to the user interface; a first associating part 1202 configured to Touch operation information and screen element information to obtain the degree of relevance between the screen element information and multiple applications, and the multiple applications are applications of the same type; the first display part 1203 is configured to recommend and display applications whose degree of relevance meets a preset condition .
  • the touch operation information includes: touch operation and operation content
  • the first obtaining part 1201 is further configured to obtain a touch operation on the user interface; and display operation content corresponding to the touch operation.
  • the touch operation information includes: touch operation and operation content
  • the first acquiring part 1201 is further configured to acquire touch operations acting on the user interface; based on the touch operations, acquire screen element information corresponding to the user interface; Based on the touch operation and screen element information, the corresponding operation content is displayed.
  • the first associating part 1202 is further configured to obtain the degree of association between the screen element information corresponding to the user interface and multiple applications in response to the selection of the operation content.
  • the touch operation information includes: touch operation and operation content
  • the first acquisition part 1201 is also configured to display the operation content corresponding to the selected content under the effect of the selection operation on the user interface; within the preset time range, acquire the touch operation on the operation content; the touch operation includes selecting operate.
  • the first associating part 1202 is further configured to associate with preset sample fusion information and preset sample sequence behavior information of each application on the basis of screen element information and touch operation information , so as to obtain the degree of association between the screen element information and each application of the plurality of applications.
  • the first association part 1202 includes a first screen element hidden information encoding part, a first touch action hidden information encoding part and a first association degree processing part;
  • the first screen element hidden information encoding part is configured to encode the screen element information to obtain the screen element hidden information
  • the first touch action hidden information encoding part is configured to encode touch operation information to obtain touch action hidden information
  • the first association degree processing part is configured to associate each application based on screen element hidden information, touch action hidden information, sample fusion information, and sample sequence behavior information of each application to obtain the association degree of each application.
  • the screen element information includes: structural knowledge corresponding to the user interface
  • the first screen element hidden information encoding part is further configured to encode the structural knowledge to obtain screen element hidden information.
  • the screen element information includes: the structural knowledge, textual knowledge, visual knowledge corresponding to the user interface, and the current application service set related to the user interface; the current application service set is determined by analyzing the service of the user interface ;
  • the hidden information encoding part of the first screen element is also configured to encode structural knowledge, textual knowledge, and visual knowledge respectively to obtain structural information, textual information, and visual information; to encode the current application service set to obtain service bias configuration information; splicing at least one of text information, visual information, service bias information, and structural information to obtain screen element hidden information.
  • structural knowledge includes: at least one of control type, control function, and control structure distribution;
  • textual knowledge includes: at least one of character recognition information and special symbol recognition information, and text original information;
  • visual The type knowledge includes: at least one of image content identification information and image source information, and original image information.
  • the hidden information encoding part of the first screen element is further configured to perform vectorized encoding on structural knowledge to obtain structural information; and perform vectorized encoding on at least one of character identification information and special symbol identification information , to obtain text content understanding coding information; and stacking and self-encoding the original text information to obtain text high-dimensional information; splicing text high-dimensional information and text content understanding coding information to obtain text information; image content identification information and image sources Informatization encoding of at least one of the information to obtain image content understanding encoding information; and stacking and self-encoding the original image information to obtain image high-dimensional information; splicing image high-dimensional information and image content understanding encoding information to obtain visual information.
  • the touch operation information includes: operation content and touch operation
  • the first touch action hidden information encoding part is further configured to encode the touch operation and operation content respectively to obtain touch operation information and operation content information; splicing the touch operation information and operation content information to obtain touch action hidden information .
  • the first acquisition part 1201 is further configured to identify control information on the user interface or the area to be identified to obtain structural knowledge; to identify text content on the user interface or the area to be identified to obtain textual knowledge; Perform image content recognition on the user interface or the area to be recognized to obtain visual knowledge; perform service analysis on the user interface to determine the current application service set; at least one of textual knowledge, visual knowledge, and the current application service set, as well as the structure type knowledge as screen element information.
  • the first obtaining part 1201 is further configured to obtain the structure information of the control tree corresponding to the user interface or the region to be identified through the accessibility service interface; perform control information identification on the structure information of the control tree to obtain the structure type knowledge.
  • the first association part 1202 further includes a first collaborative joint learning part (corresponding to the aforementioned collaborative joint learning module);
  • the first collaborative joint learning part is configured to perform collaborative joint learning on screen element hidden information and touch action hidden information to obtain fusion information to be analyzed;
  • the first obtaining part 1201 is further configured to obtain historical sequence behavior information of each application determined based on historical behavior data;
  • the first correlation degree processing part is also configured to perform similarity processing based on the historical sequence behavior information and sample sequence behavior information of each application, and to perform similarity processing based on the fusion information to be analyzed and the sample fusion information of each training sample, and obtain each Application similarity.
  • the first association degree processing part is further configured to determine historical behavior parameters of multiple applications based on historical behavior data; for each application, set the historical behavior parameters corresponding to the application as preset parameters , combining the historical behavior parameters of multiple applications except the application, to obtain the historical sequence behavior information of the application.
  • the first correlation processing part is further configured to perform similarity processing based on the historical sequence behavior information and sample sequence behavior information of each application, and determine the first similarity information of each application for each training sample; Perform similarity processing on the fusion information to be analyzed and the sample fusion information of each training sample to obtain the second similarity information corresponding to each training sample; based on the first similarity information of each application for each training sample, and the corresponding The second similarity information is to obtain the similarity of each application.
  • the first association degree processing part is further configured to obtain, based on the first similarity information of each application for each training sample and the second similarity information corresponding to each training sample, each application for each training sample The third similarity degree of each application; for each application, the third similarity degree of each training sample is summed to obtain the similarity degree of each application.
  • the application recommending device 120 further includes a training part, the training part is configured to determine the sample screen element hidden information, the sample touch action hidden information and the sample service information corresponding to each training sample according to the preset training data. ; Perform cooperative joint learning on the hidden information of the sample screen elements and the hidden information of the sample touch actions to obtain the preset sample fusion information of each training sample; determine the preset sample sequence behavior of each application according to the sample service information of each training sample information.
  • the first display part 1203 is further configured to recommend and display the application with the highest similarity of each application on the user interface; or, on the user interface, recommend to display the preset number of applications with the highest similarity to each application. application.
  • an application includes: an application type, and/or, an application service.
  • the first display part 1203 is further configured to generate service prompt information based on the application meeting the preset condition; on the user interface, it is recommended to display the service prompt information.
  • the first obtaining part 1201 is further configured to obtain the screen element information of the full interface corresponding to the user interface when the touch operation information is satisfied, indicating that the operation content meets the preset trigger recommendation condition;
  • the trigger recommendation condition represents the expected operation intention.
  • the first acquisition part 1201 is further configured to determine the current area to be identified at the touch position where the touch operation information acts; identify the current area to be identified, and determine screen element information.
  • the application recommending device 120 includes an identification part, the identification part is configured to determine the next area to be identified under the condition that no screen element information is identified in the current area to be identified, and perform Recognize until the screen element information is obtained or the complete interface is recognized; the next region to be recognized is larger than the current region to be recognized.
  • the embodiment of the application also provides another application recommendation device, as shown in Figure 13, which is a schematic structural diagram of an application recommendation device provided in the embodiment of the application, the application
  • the recommending device 130 includes: a second acquiring part 1301 configured to acquire touch operation information for the user interface and screen element information corresponding to the user interface; a second associating part 1302 configured to The operation information is jointly learned, and associated with the preset sample fusion information and the preset sample sequence behavior information of each application, and the degree of correlation between the screen element information and multiple applications is obtained; the second display part 1303 is configured as The recommended display is performed on the applications whose degree of relevance meets the preset condition.
  • the second association part 1302 includes a second screen element latent information coding part, a second touch action hidden information coding part, a second cooperative joint learning part (corresponding to the above collaborative joint learning module) and a second association degree processing part;
  • the second screen element hidden information encoding part is configured to encode the screen element information to obtain the screen element hidden information
  • the second touch action hidden information encoding part is configured to encode touch operation information to obtain touch action hidden information
  • the second collaborative joint learning part is configured to perform collaborative joint learning on screen element hidden information and touch action hidden information to obtain fusion information to be analyzed;
  • the second obtaining part 1301 is also configured to obtain the historical sequence behavior information of each application determined based on the historical behavior data;
  • the second correlation processing part is configured to perform similarity processing based on the historical sequence behavior information and sample sequence behavior information of each application, and to perform similarity processing based on the fusion information to be analyzed and the sample fusion information of each training sample, and obtain each application similarity.
  • the second correlation degree processing part is configured to perform similarity processing based on the historical sequence behavior information and sample sequence behavior information of each application, and determine the first similarity information of each application for each training sample; Analyzing the fusion information and the sample fusion information of each training sample to perform similarity processing to obtain the second similarity information corresponding to each training sample; 2. Similarity information to obtain the similarity of each application.
  • the application recommendation device when the application recommendation device provided in the above embodiment performs application recommendation, it only uses the division of the above-mentioned program parts for illustration. That is, the internal structure of the device is divided into different program parts to complete all or part of the processing described above.
  • the application recommending device provided by the above embodiment is based on the same idea as the application recommending method embodiment, and its specific implementation process and beneficial effects are detailed in the method embodiment, and will not be repeated here.
  • the technical details not disclosed in the device embodiment of this application please refer to the description of the method embodiment of this application for understanding.
  • FIG. 14 is a schematic diagram of the composition and structure of the application recommendation device proposed in the embodiment of the present application.
  • the application recommendation device 140 may also include a communication interface 1403 and a bus 1404.
  • the bus 1404 is configured to connect the processor 1401, the memory 1402, and the communication interface 1403 .
  • the bus 1404 is configured to connect the communication interface 1403, the processor 1401, and the memory 1402, and communicate with each other among these devices.
  • the above-mentioned processor 1401 is configured to obtain touch operation information acting on the user interface; obtain screen element information corresponding to the user interface based on the touch operation information; , to obtain the degree of association between the screen element information and multiple applications, and the multiple applications are of the same type; and recommend and display the applications whose degree of association satisfies a preset condition.
  • the above-mentioned processor 1401 is further configured to obtain touch operation information for the user interface, and obtain screen element information corresponding to the user interface; perform joint learning based on the screen element information and touch operation information, and Perform association processing with the preset sample fusion information and the preset sample sequence behavior information of each application to obtain the degree of association between the screen element information and multiple applications, and the multiple applications are the same type of application;
  • the application displays recommendations.
  • the above-mentioned processor 1401 may be an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a digital signal processor (Digital Signal Processor, DSP), a digital signal processing device (Digital Signal Processing Device, DSPD) , Programmable Logic Device (ProgRAMmable Logic Device, PLD), Field Programmable Gate Array (Field ProgRAMmable Gate Array, FPGA), Central Processing Unit (Central Processing Unit, CPU), controller, microcontroller, microprocessor in at least one.
  • ASIC Application Specific Integrated Circuit
  • DSP Digital Signal Processor
  • DSPD Digital Signal Processing Device
  • PLD Programmable Logic Device
  • Field Programmable Gate Array Field ProgRAMmable Gate Array
  • CPU Central Processing Unit
  • controller microcontroller, microprocessor in at least one.
  • the memory 1402 in the application recommendation device 140 may be connected to the processor 1401, and the memory 1402 is configured to store executable program codes and data, where the program codes include computer operation instructions.
  • Each of the memories 1402 may include a high-speed RAM memory, and may also include a non-volatile memory, for example, at least two disk memories.
  • the above-mentioned memory 1402 can be a volatile memory (volatile memory), such as a random access memory (Random-Access Memory, RAM); or a non-volatile memory (non-volatile memory), such as a read-only Memory (Read-Only Memory, ROM), flash memory (flash memory), hard disk (Hard Disk Drive, HDD) or solid-state drive (Solid-State Drive, SSD); or a combination of the above types of memory, and send to the processor 1401 provides instructions and data.
  • volatile memory such as a random access memory (Random-Access Memory, RAM); or a non-volatile memory (non-volatile memory), such as a read-only Memory (Read-Only Memory, ROM), flash memory (flash memory), hard disk (Hard Disk Drive, HDD) or solid-state drive (Solid-State Drive, SSD); or a combination of the above types of memory, and send to the processor 1401 provides instructions and data.
  • ROM read-only Memory
  • each functional part in this embodiment may be integrated into one processing part, each part may exist separately physically, or two or more parts may be integrated into one part.
  • the above-mentioned integrated part can be implemented not only in the form of hardware, but also in the form of software function part.
  • the integrated part is realized in the form of a software function part and is not sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of this embodiment is essentially or The part contributed by the prior art or the whole or part of the technical solution can be embodied in the form of software products, the computer software products are stored in a storage medium, and include several instructions to make a computer device (which can be a personal A computer, a server, or a network device, etc.) or a processor (processor) executes all or part of the steps of the method of this embodiment.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read only memory (Read Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other various media that can store program codes.
  • An embodiment of the present application provides a computer-readable storage medium on which a program is stored, and when the program is executed by a processor, the application recommendation method in any one of the above embodiments is implemented.
  • the program instruction corresponding to an application recommendation method in this embodiment may be stored on a storage medium such as an optical disk, a hard disk, or a USB flash drive.
  • the program instruction corresponding to an application recommendation method in the storage medium is When the electronic device reads or is executed, the application recommendation method in any one of the above embodiments can be implemented.
  • the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) having computer-usable program code embodied therein.
  • a computer-usable storage media including but not limited to disk storage and optical storage, etc.
  • These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions
  • the device realizes the function specified in implementing one or more procedures of the flowchart and/or one or more blocks of the block diagram.
  • the embodiment of the present application discloses an application recommendation method, device, equipment and computer-readable storage medium.
  • the method includes: obtaining touch operation information acting on the user interface; obtaining screen element information corresponding to the user interface based on the touch operation information; both the screen element information and the touch operation information are interactive data related to the current operation of the user, and can Meet the immediate service needs of users.
  • the degree of correlation between the screen element information and multiple applications is obtained, and the multiple applications are of the same type; and the applications whose correlation degrees meet the preset conditions are recommended and displayed.
  • the embodiment of the present application considers the interaction data of the user interface from two dimensions by combining the dual mode information of the screen element information and the touch operation information, and then combines the degree of association with multiple applications to improve the accuracy of the application recommendation results. sex.

Abstract

本申请实施例公开了一种应用推荐方法、装置、设备和计算机可读存储介质。该方法包括:获取作用于用户界面的触控操作信息;基于触控操作信息,获取用户界面对应的屏幕元素信息;屏幕元素信息和触控操作信息均是与用户当前操作相关的交互数据,能够满足用户的即时服务需求。基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;对关联度满足预设条件的应用进行推荐显示。本申请实施例通过结合屏幕元素信息和触控操作信息的双重模态信息,从两个维度对用户界面的交互数据进行考虑,然后结合与多个应用的关联度,提高了应用推荐结果的准确性。

Description

应用推荐方法、装置、设备和计算机可读存储介质
相关申请的交叉引用
本申请基于申请号为202111387371.6、申请日为2021年11月22日、申请名称为“应用推荐方法、装置、设备和计算机可读存储介质”的中国专利申请提出,并要求该中国专利申请的优先权,该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本申请涉及计算机技术领域,尤其涉及一种应用推荐方法、装置、设备和计算机可读存储介质。
背景技术
随着智能交互技术的不断发展,以智能设备为中心的交互模式,逐渐向以人为中心的服务交互模式发展。通过在智能设备上向用户进行智能推荐服务,提高了用户操作的便捷性。
相关技术中,通过采集用户对智能设备的历史行为数据,并结合热门服务使用数据,生成应用推荐结果。
然而,相关技术中依赖于用户的历史行为数据和热门服务使用数据,使得用户在进行当前操作,实现与屏幕的交互时,只能被动接收应用推荐,应用推荐结果往往与当前操作没有关联,不能满足用户的即时服务需求,降低了应用推荐结果的准确性。
发明内容
本申请实施例期望提供一种应用推荐方法、装置、设备和计算机可读存储介质,通过结合屏幕元素信息和触控操作信息的双重模态信息,从两个维度对用户界面的交互数据进行考虑,然后结合与多个应用的关联度,提高了应用推荐结果的准确性。
第一方面,本申请实施例提供一种应用推荐方法,所述方法包括:获取作用于用户界面的触控操作信息;基于所述触控操作信息,获取所述用户界面对应的屏幕元素信息;基于所述触控操作信息和所述屏幕元素信息,获取所述屏幕元素信息与多个应用的关联度,所述多个应用为同一类型应用;对所述关联度满足预设条件的应用进行推荐显示。
第二方面,本申请实施例提供一种应用推荐方法,所述方法包括:获取针对用户界面的触控操作信息,以及获取所述用户界面对应的屏幕元素信息;基于所述屏幕元素信息和所述触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取所述屏幕元素信息与多个应用的关联度;对所述关联度满足预设条件的应用进行推荐显示。
第三方面,本申请实施例提供一种应用推荐装置,所述装置包括:第一获取部分,被配置为获取作用于用户界面的触控操作信息;以及基于所述触控操作信息,获取所述用户界面对应的屏幕元素信息;第一关联部分,被配置为基于所述触控操作信息和所述屏幕元素信息,获取所述屏幕元素信息与多个应用的关联度,所述多个应用为同一类型应用;第一显示部分,被配置为对所述关联度满足预设条件的应用进行推荐显示。
第四方面,本申请实施例提供一种应用推荐装置,所述装置包括:第二获取部分,被配置为获取针对用户界面的触控操作信息,以及获取所述用户界面对应的屏幕元素信息;第二关联部分,被配置为基于所述屏幕元素信息和所述触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取所述屏幕元素信息与多个应用的关联度;第二显示部分,被配置为对所述关联度满足预设条件的应用进行推荐显示。
第五方面,本申请实施例提供一种应用推荐设备,所述应用推荐设备包括存储器和处理器;所述存储器存储有可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序时,实现上述第一方面或第二方面所述的应用推荐方法。
第六方面,本申请实施例提供一种计算机可读存储介质,其上存储有可执行指令,被配置为被处理器执行时,实现上述第一方面或第二方面所述的应用推荐方法。
本申请实施例提供了一种应用推荐方法、装置、设备和计算机可读存储介质。根据本申请实施例提供的方案,获取作用于用户界面的触控操作信息;基于触控操作信息,获取用户界面对应的屏幕元素信息;屏幕元素信息和触控操作信息均是与用户当前操作相关的交互数据,能够满足用户的即时服务需求。基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;对关联度满足预设条件的应用进行推荐显示。本申请实施例通过结合屏幕元素信息和触控操作信息的双重模态信息,从两个维度对用户界面的交互数据进行考虑,然后结合与多个应用的关联度,提高了应用推荐结果的准确性。
附图说明
图1为本申请实施例提供的一种应用推荐结果的示例性的示意图;
图2为本申请实施例提供的一种应用推荐方法的可选的步骤流程图;
图3为本申请实施例提供的另一种应用推荐方法的可选的步骤流程图;
图4为本申请实施例提供的再一种应用推荐方法的可选的步骤流程图;
图5为本申请实施例提供的一种计算屏幕元素隐向量的可选的流程图;
图6为本申请实施例提供的一种计算触控动作隐向量的可选的流程图;
图7为本申请实施例提供的又一种应用推荐方法的可选的步骤流程图;
图8为本申请实施例提供的一种生成推荐结果的可选的流程图;
图9为本申请实施例提供的一种应用推荐结果的示例性的表现形式示意图;
图10为本申请实施例提供的另一种应用推荐结果的示例性的表现形式示意图;
图11为本申请实施例提供的又一种应用推荐方法的可选的步骤流程图;
图12为本申请实施例提供的一种应用推荐装置的结构示意图;
图13为本申请实施例提供的另一种应用推荐装置的结构示意图;
图14为本申请实施例提供的一种应用推荐设备组成结构示意图。
具体实施方式
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述。应当理解的是,此处所描述的一些实施例仅仅用以解释本申请的技术方案,并不用于限定本申请的技术范围。
为了更好地理解本申请实施例中提供的应用推荐方法,在对本申请实施例的技术方案进行介绍之前,先对相关技术进行说明。
相关技术中,智能应用推荐设备通过采集用户对智能设备的历史行为数据,并结合热门使用服务数据,生成应用推荐结果,如图1所示,图1为本申请实施例提供的一种应用推荐结果的示例性的示意图。图1以终端是手机为例,展示手机屏幕的应用推荐场景,图1中的“应用建议”和“热门应用”栏目中所展示的内容,属于终端的智能应用推荐设备通过计算分析得到的应用推荐结果。其中,“应用建议”是根据采集到的用户对智能设备的历史行为数据生成的应用推荐,即基于用户历史行为的服务推荐,“热门应用”是根据智能设备联网到的热门使用服务数据生成的应用推荐,即基于热门服务的服务推荐,在图1中分别以不同的字母表示不同的推荐内容。
智能应用推荐方法包括以下特点:在推荐数据的使用上,通常采用用户的历史浏览、点击等交互数据,以及互联网热门数据,作为对用户进行应用推荐时的参考依据。在推荐服务的内容上,主要为用户提供应用推荐,广告推荐或资讯类消息等推荐。在推荐服务的性质上,主要是为了提高应用软件的点击率与活跃率,因此,侧重于引导用户使用智能应用推荐设备想要让用户使用的应用软件。
基于相关技术中的上述特征,智能应用推荐在用户服务需求、产品表现上至少存在以下一个或多个缺点:(1)由于相关技术中依赖于用户的历史行为数据和热门使用服务数据,对用户即将要使用的服务进行推荐,难以覆盖到用户即时的服务需求。(2)由于相关技术中只是使用了用户的历史行为数据,并不考虑用户的主动意图操作,也就是说,智能应用推荐设备与用户的主动行为之间无交互,用户只能被动的接收应用推荐。(3)由于相关技术中仅使用了用户的历史行为数据,信息模态单一,从而降低了相关技术中应用推荐结果的准确性。
本申请实施例提供一种应用推荐方法,如图2所示,图2为本申请实施例提供的一种应用推荐方法 的步骤流程图,应用推荐方法包括以下步骤:
S201、获取作用于用户界面的触控操作信息。
在本申请实施例中,用户与应用推荐设备进行交互时,用户在用户界面上对其感兴趣的信息进行操作,操作过程中产生触控操作信息。该触控操作信息是作用于用户界面的信息,应用推荐设备可以获取到该触控操作信息。
在本申请实施例中,用户界面为用户与应用推荐设备进行交互时的页面,用户界面可以理解为当前页面、当前用户界面。触控操作信息表征用户对于用户界面之间的交互数据。触控操作信息可以包括:操作内容和触摸操作。触摸操作包括但不限于点击、双击、长按、双指捏合、双指方法、滑动和拖拽;操作内容包括但不限于复制、转发、分享、拖拽、提醒、保存、收藏和停留等。其中,停留指的是用户对用户界面上的控件进行操作,且操作内容为除其他用户操作的所有内容,以操作系统弹出的子菜单中的可选控件包括复制、转发、分享、拖拽、提醒、保存和收藏为例进行说明,用户未对子菜单中的可选控件进行选择,而是选择了其他控件,或者用户选中了除子菜单之外的控件。
在一些实施例中,触控操作信息包括:触摸操作和操作内容;上述S201可以通过以下三个示例实现。第一个示例,获取作用于用户界面的触摸操作;显示与触摸操作对应的操作内容。
在本申请实施例中,在获取触控操作信息时,是由用户的触摸操作触发的,先获取作用于用户界面的触摸操作,在用户界面中显示与触摸操作对应的操作内容。示例性的,以触摸操作是触控手势为例进行说明,用户需要对某一文本信息进行复制操作,当检测到用户的触控手势是“长按”时,生成与“长按”对应的操作内容,操作内容是复制、分享、转发、保存、收藏等,用户选择操作内容“复制”。
在本申请实施例中,获取作用于用户界面的触摸操作,显示与触摸操作对应的操作内容,若用户对操作内容进行选择,则获取操作内容,从而获得触控操作信息,相较于直接获取触控操作信息的方案,减少了资源消耗。
在一些实施例中,触控操作信息包括:触摸操作和操作内容;上述S201通过以下第二个示例实现。第二个示例,获取作用于用户界面的触摸操作;基于触摸操作,获取用户界面对应的屏幕元素信息;基于触摸操作和屏幕元素信息,显示对应的操作内容。
在本申请实施例中,在获取触控操作信息时,是由用户的触摸操作触发的,触摸操作反映用户需要对用户界面中的某些内容进行操作,以便接下来选择操作内容,屏幕元素信息反映能够支持的操作内容。因此,显示的操作内容是与触摸操作和屏幕元素信息相关的内容,基于触摸操作和屏幕元素信息,显示对应的操作内容,提高了操作内容的准确性。示例性的,以触摸操作是触控手势为例进行说明,用户需要对某一文本信息进行复制操作,当检测到用户的触控手势“长按”时,获取用户界面对应的屏幕元素信息,基于触控手势所在的位置和屏幕元素信息生成操作内容,操作内容是复制、分享、转发、保存、收藏等,用户选择操作内容“复制”。
在本申请实施例中,先获取作用于用户界面的触摸操作,基于触摸操作,获取用户界面对应的屏幕元素信息;基于触摸操作和屏幕元素信息,显示对应的操作内容,若用户对操作内容进行选择,则获取操作内容,从而获得触控操作信息,相较于直接获取触控操作信息的方案,减少了资源消耗。
在一些实施例中,触控操作信息包括:触摸操作和操作内容;上述S201通过以下第三个示例实现。第三个示例,在针对用户界面的选择操作的作用下,显示被选中的内容对应的操作内容;在预设时间范围内,获取针对操作内容的触摸操作;触摸操作包括选择操作。
在本申请实施例中,用户与应用推荐设备进行交互的过程中,操作系统会一直产生数据,包括触控操作信息和屏幕元素信息,本申请实施例并不是一直获取触控操作信息和屏幕元素信息,也不会一直向用户进行应用推荐,操作系统也不会一直获取触摸操作和屏幕元素信息。可以理解为,用户可以一直对应用推荐设备进行触摸操作,但是应用推荐设备并不是响应所有的触摸操作。在针对用户界面的选择操作的作用下,显示被选中的内容对应的操作内容,例如被选中的操作内容是复制、转发、分享、拖拽、提醒、保存和收藏中的至少一项时,检测到用户具有表征想要超级分享、转发或打开等的强意图,才会在该选择操作的前后一段时间范围内,获取针对操作内容的触摸操作。同时也说明了本申请实施例的应用推荐方法是由操作内容触发的应用推荐。
在本申请实施例中,在针对用户界面的选择操作的作用下,显示被选中的内容对应的操作内容;从而在预设时间范围内,获取针对操作内容的触摸操作,相较于一直获取触摸操作的方法,减少了资源消耗。
S202、基于触控操作信息,获取用户界面对应的屏幕元素信息。
在本申请实施例中,随着用户与用户界面之间的实时交互,用户界面对应屏幕元素信息是变化的。操作系统会一直产生数据,包括触控操作信息和屏幕元素信息,应用推荐设备并不是一直向用户进行应用推荐。基于触控操作信息,获取用户界面对应的屏幕元素信息,该屏幕元素信息是与用户触控操作信 息对应的屏幕元素,提高了屏幕元素信息的准确性。
在本申请实施例中,屏幕元素信息表征用户界面中显示的所有元素的信息,包括但不限于控件信息、文本类信息和图像类信息。
在一些实施例中,上述S202中在获取用户界面对应的屏幕元素信息时,可以通过以下方式实现。对用户界面或待识别区域进行控件信息识别,得到结构型知识;对用户界面或待识别区域进行文字内容识别,得到文本型知识;对用户界面或待识别区域进行图像内容识别,得到视觉型知识;对用户界面进行服务解析,确定当前应用服务集合;文本型知识、视觉型知识和当前应用服务集合中的至少一种,以及结构型知识作为屏幕元素信息。
在本申请实施例中,结构型知识表征实体和实体之间的联系,也可以理解为控件和控件之间的联系,文本型知识表征电影、文字或链接之类的文本内容,视觉型知识表征封面或海报之类的图像内容。
示例性的,通过对当前用户界面进行服务解析确定当前应用服务集合,可以理解为选择与当前用户界面相关的应用,作为当前应用服务集合,减少需要编码的应用的数量,从而减少资源消耗。
需要说明的是,待识别区域可以是由本领域技术人员根据实际需求,在用户界面对应的全部区域中选取的其中部分区域,也可以是根据用户操作区域确定待识别区域,从而提高待识别区域的准确性,对此本申请实施例不作限制。
在本申请实施例中,对于任意一次应用推荐过程,在获取用户界面对应的屏幕元素信息时,均需要上述的控件信息识别,但是由于有些用户界面中不存在图像和/或文本,因此,在进行解析时,不一定需要文字内容识别和图像内容识别,也就是用户界面中不一定包括文本型知识、视觉型知识和当前应用服务集合,因此,将文本型知识、视觉型知识和当前应用服务集合中的至少一种,以及结构型知识作为屏幕元素信息,提高了屏幕元素信息的丰富性和准确性。
在一些实施例中,在对用户界面或待识别区域进行控件信息识别,得到结构型知识时,可以包括以下步骤:通过无障碍服务接口,获取用户界面或待识别区域对应的控件树的结构信息;对控件树的结构信息进行控件信息识别,得到结构型知识。
在本申请实施例中,无障碍服务(Accessibility Service)是一套可以模拟操作的系统级别的应用程序接口(Application Programming Interface,API)。用户同意应用获取无障碍服务的权限之后就可以模拟操作,依次控制用户的应用推荐设备。无障碍服务可以用于抢红包、自动回复、一键获取权限等应用程序中,实现一键操作。示例性的,控件信息的识别方法可以采用操作系统提供的无障碍服务接口,获取用户界面或待识别区域对应的整个控件树的结构信息,在用户界面或待识别区域刷新时,可以通过调用无障碍服务接口,获取更新后的界面的整个控件树的结构。然后根据整个控件树的结构信息,通过拓扑技术对控件树的结构信息进行控件信息识别,得到结构型知识。
在本申请实施例中,通过无障碍服务接口,获取用户界面或待识别区域对应的控件树的结构信息;对控件树的结构信息进行控件信息识别,得到结构型知识,提高了结构型知识的准确性。
S203、基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用。
在本申请实施例中,应用推荐设备是为了满足用户即时的服务需求,因此需要获得用户界面的触控操作信息以及用户界面对应的屏幕元素信息。屏幕元素信息和触控动作信息反映了与用户即时需求相关的强意图,可以理解为用户相关的即时信息,在获取即时信息之后,需要计算屏幕元素信息与多个应用之间的关联度。
在本申请实施例中,多个应用可以为同一类型应用。同一类型的应用可以理解为应用所对应功能的表现形式为同一类型,例如,用于向用户播放图像或视频的多个不同的播放软件,为同一类型的应用;用于进行购物的多个不同的购物软件,为同一类型的应用;用于进行社交的多个不同的社交软件,为同一类型的应用。
示例性的,基于触控操作信息和屏幕元素信息可以确定多个应用,多个应用为同一类型的应用,该类型的应用均能够支持当前用户界面中触控操作信息中的操作内容和屏幕元素信息,然后将屏幕元素信息与多个应用之间进行协同联合学习和相似度处理,得到多个关联度。
在本申请实施例中,协同联合学习(collaborative joint learning)是将屏幕元素隐信息、触控动作隐信息进行融合的过程,相似度处理是将融合后的信息与协同过滤(collaborative filtering)相结合的过程。以产品推荐为例进行说明,协同过滤算法是基于对用户历史行为数据的挖掘发现用户的喜好偏向,并预测用户可能喜好的产品进行推荐。可以理解为“猜你喜欢”,和“购买了该商品的人也喜欢”等功能。可以通过以下方式实现:根据和你有共同喜好的人给你推荐、根据你喜欢的物品给你推荐相似物品,以及根据以上条件综合推荐。
在一些实施例中,应用包括:应用类型,和/或,应用服务。
在本申请实施例中,应用的表现形式有很多,应用可以是应用类型、应用服务,或者应用类型和应用服务的结合,应用类型可以理解为应用软件的类型,如图1中列出的应用建议以及热门应用的各个不同的应用服务。应用服务可以理解为在应用软件中对用户界面中的文本或图像进行展示,例如,在播放软件中打开图像、视频等文件,在购物软件中打开该购物链接,在浏览器中打开普通链接,在拨号软件中向其他用户拨打电话,在办公软件中打开文档。不同用户在面对每条交互数据时会产生不同的选择结果,例如,选择不同类型的应用软件;选择在应用软件中打开图像或文本;或者,先选中不同类型的应用软件,再进行图像或文本的打开。对应的,应用推荐结果也可以是应用类型和应用服务的结合。本申请实施例中的应用推荐也可以理解为服务推荐、应用服务推荐、应用软件推荐等,对此本申请实施例不作限制。
在本申请实施例中,应用或应用推荐结果包括应用类型,和/或,应用服务,提高了应用推荐结果的丰富性。
在本申请实施例中,触控操作信息和屏幕元素信息与当前用户界面相关,能够满足用户的即时需求,基于触控操作信息和屏幕元素信息,确定多个同一类型的应用,相较于直接获取屏幕元素信息与所有应用的关联度,获取屏幕元素信息与多个同一类型的应用的关联度,减少了资源消耗,提高了关联度的准确性。
S204、对关联度满足预设条件的应用进行推荐显示。
在本申请实施例中,将S203得到的多个关联度中,满足预设条件的关联度对应的应用作为应用推荐结果,并对该应用进行推荐显示。示例性的,先向用户显示是否需要打开应用的标志,该标志可以是以窗口、文本、图画、文件夹等形式示出,以窗口为例,若用户点击窗口,则显示关联度满足预设条件的应用。也可以直接向用户显示关联度满足预设条件的应用,用户直接对应用进行选择即可。
在本申请实施例中,预设条件可以由本领域技术人员根据实际需求进行适当设置,例如,若预设条件为关联度大于预设阈值,则向用户推荐一个或多个应用,若预设条件为关联度最大值,则向用户推荐一个应用,对此本申请实施例不作限制。
在一些实施例中,上述S204可以通过以下方式实现。在用户界面,推荐显示各个应用的相似度最高的应用;或者,在用户界面,推荐显示各个应用的相似度最高的预设数量的应用。
在本申请实施例中,向用户界面推荐显示相似度最高的应用,示例性的,直接进入该应用,不需要用户进行再次操作即可跳转到该应用,提高了应用推荐效率。或者,将该应用在用户界面上推荐给用户,用户只用选择是否打开或进入该应用,不需要用户在多个应用中进行选择,提高了应用推荐效率。
在本申请实施例中,推荐显示各个应用的相似度最高的预设数量的应用,向用户推荐预设数量的应用,用户可以选择与自身需求强相关的应用,提高了应用推荐的多样性。需要说明的是,预设数量可以由本领域技术人员根据实际情况进行适当设置,例如,2个、3个、4个等,对此本申请实施例不作限制。
根据本申请实施例提供的方案,获取作用于用户界面的触控操作信息;基于触控操作信息,获取用户界面对应的屏幕元素信息;屏幕元素信息和触控操作信息均是与用户当前操作相关的交互数据,能够满足用户的即时服务需求。基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;对关联度满足预设条件的应用进行推荐显示。本申请实施例通过结合屏幕元素信息和触控操作信息的双重模态信息,从两个维度对用户界面的交互数据进行考虑,然后结合与多个应用的关联度,提高了应用推荐结果的准确性。
在一些实施例中,触控操作信息包括操作内容;获取用户界面对应的屏幕元素信息;上述S202可以通过以下两个示例实现。第一个示例,在满足触控操作信息表征其操作内容满足预设触发推荐条件的情况下,获取用户界面对应的全界面的屏幕元素信息;预设触发推荐条件表征预期的操作意图。
在本申请实施例中,用户与应用推荐设备进行实时交互的过程中,应用推荐设备并不是一直向用户进行应用推荐。在满足触控操作信息表征其操作内容未达到预期的操作意图的情况下,表示用户没有强意图,此时并不会对用户进行应用推荐,不需要获取用户界面对应的全界面的屏幕元素信息。在满足触控操作信息表征其操作内容达到预期的操作意图的情况下,表示用户具有表征想要超级分享、转发或打开等强意图,则获取用户界面对应的全界面的屏幕元素信息。
需要说明的是,用户界面对应的全界面可以理解为用户界面的全部区域,预设触发推荐条件可以由本领域技术人员根据实际需求进行适当设置,能够对用户是否具有强意图进行区分即可。
在一些实施例中,预设触发推荐条件表征预期的操作意图,示例的,预设触发条件可以是预设的操作内容,预设的操作内容可以是复制、转发、分享、拖拽、提醒、保存或收藏,即,预设触发推荐条件可以为以下至少一个:复制、转发、分享、拖拽、提醒、保存和收藏,但对此本申请实施例不作限制。当触控操作信息的操作内容或其所对应的功能,满足预设触发推荐条件时,获取用户界面对应的全界面 的屏幕元素信息。
例如,示例性的,预设触发条件是操作内容中的“复制”,用户对于某一链接的触摸操作是“长按”,然后操作系统弹出子菜单,用户操作内容是选择了子菜单中的“复制”,表示用户具有表征想要超级分享、转发或打开等的强意图,说明用户的操作内容满足预设触发推荐条件。
在本申请实施例中,通过设置预设触发条件,相较于直接获取用户界面对应的全界面的屏幕元素信息的方式,减少了资源消耗。
在一些实施例中,上述S202通过以下第二个示例实现。第二个示例,在触控操作信息作用的触控位置,确定当前待识别区域;对当前待识别区域进行识别,确定屏幕元素信息。
在本申请实施例中,屏幕元素信息是针对用户界面中显示的各个不同类型的元素的信息,触控操作信息作用的触控位置是用户界面中的其中部分界面,也就是说用户触摸操作针对的并不是用户界面的全部区域,因此,根据触控操作信息作用的触控位置,确定当前待识别区域,当前待识别区域的面积小于用户界面的全部区域,然后对当前待识别区域进行识别,确定待识别区域中屏幕元素信息。
在本申请实施例中,根据触控操作信息作用的触控位置确定当前待识别区域,然后对当前待识别区域进行识别,确定屏幕元素信息,不需要获取用户界面的全部区域的屏幕元素信息,减少了数据处理量,提高了应用推荐效率。
在一些实施例中,上述第二个示例中在触控操作信息作用的触控位置,确定当前待识别区域之后,该应用推荐方法还包括以下步骤:在满足当前待识别区域中未识别出屏幕元素信息的情况下,确定下一个待识别区域,在下一个待识别区域进行识别,直至得到存在屏幕元素信息或者识别完全界面时为止;下一个待识别区域大于当前待识别区域。
在本申请实施例中,以触控操作信息作用的触控位置为中心,将预设范围内的区域确定为当前待识别区域,在满足当前待识别区域中未识别出屏幕元素信息的情况下,说明需要进一步扩大待识别区域的面积,例如,扩大预设范围,或者,以当前待识别区域为基准,向某个方向或四周进行扩展,从而确定下一个待识别区域,下一个待识别区域大于当前待识别区域。然后在下一个待识别区域进行识别,直至得到存在屏幕元素信息或者识别完全界面时为止。
在本申请实施例中,根据是否识别出屏幕元素信息的结果,判断是否需要对待识别区域进行范围扩大,直至得到屏幕元素信息,通过循序渐进的方式确定待识别区域,提高了待识别区域的准确性,当得到屏幕元素信息之后,不再对待识别区域进行扩大,相交于直接对用户界面的全部区域进行识别,获取屏幕元素信息的方式,减少了资源消耗。
在一些实施例中,触控操作信息包括:触摸操作和操作内容;上述S203可以通过以下方式实现。响应于对操作内容的选择,获得用户界面对应的屏幕元素信息与多个应用的关联度。
在本申请实施例中,应用推荐设备向用户显示对应的操作内容之后,用户对操作内容进行选择,从而获得操作内容。响应于对操作内容的选择,获得用户界面对应的屏幕元素信息与多个应用的关联度。可以理解为,应用推荐方法是由操作内容触发的,在用户对操作内容进行选择之后,根据操作内容、触摸操作以及屏幕元素信息,确定多个应用,多个应用是能够支持当前用户界面中触控操作信息和屏幕元素信息的应用,从而获取屏幕元素信息与多个应用的关联度,相较于直接获取屏幕元素信息与所有应用的关联度,减少了资源消耗,提高了关联度的准确性。
在一些实施例中,上述S203还可以通过以下方式实现。基于屏幕元素信息和触控操作信息,在各个应用上与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联,从而得到屏幕元素信息与多个应用的各个应用的关联度。
在本申请实施例中,预设的样本融合信息和预设的各个应用的样本序列行为信息可以理解为预设的训练数据,预设的训练数据是在进行应用推荐之前,获取的用户与手机交互产生的数据,可以作为此次应用推荐的参考依据。预设的训练数据包括多条训练样本,对于预设的训练数据进行知识库嵌入和协同联合学习之后,得到预设的样本融合信息。预设的训练数据中还包括多个训练样本的样本服务信息,多个样本服务信息中会存在相同的应用,根据多个样本服务信息预测各个应用的偏好概率,从而得到预设的各个应用的样本序列行为信息。一个训练样本包括一条样本融合信息和该样本融合信息对应的样本服务信息,样本服务信息表征训练样本中用户对于每条交互数据(交互数据对应于样本融合信息)后续选择的应用,样本服务信息可以理解为应用真值。
在本申请实施例中,根据多条样本融合信息和每条样本融合信息对应的样本服务信息,得到各个应用的样本序列行为信息。各个应用的样本序列行为信息表征针对不同的样本融合信息,用户对各个应用的偏好概率。示例性的,第i条样本融合信息对第j个应用的样本序列行为信息,也可以理解为第i条样本融合信息对第j个应用的偏好概率。偏好概率由训练数据中样本融合信息对应的样本服务信息的数量确定,也可以理解为在同一样本融合信息下,用户对各个应用的偏好概率,偏好概率越高,样本序列 行为信息的值越大。例如,假设只存在两种应用,即,j=1或2,第i条交互数据选择了第一种应用3次,第二种应用1次,则第i条样本融合信息对第1个应用的样本序列行为信息为3/(1+3)=0.75,相应的,第i条样本融合信息对第2个应用的样本序列行为信息为0.25。对融合信息集合中的所有样本融合信息进行上述计算,可以得到各个应用的样本序列行为信息。
在本申请实施例中,基于屏幕元素信息和触控操作信息进行知识库嵌入和协同联合学习,得到待分析融合数据,然后与在各个应用上与预设的样本融合信息和预设的各个应用的样本序列行为信息进行协同联合学习和相关性处理,从而得到屏幕元素信息与多个应用的各个应用的关联度,提高了关联度的准确性。
在一些实施例中,在上述S203之前,该应用推荐方法还包括以下步骤:根据预设的训练数据,确定各个训练样本对应的样本屏幕元素隐信息、样本触控动作隐信息和样本服务信息;对样本屏幕元素隐信息和样本触控动作隐信息进行协同联合学习,得到各个训练样本的预设的样本融合信息;根据各个训练样本的样本服务信息,确定预设的各个应用的样本序列行为信息。
在本申请实施例中,预设的训练数据包括多个训练样本,一个训练样本包括样本屏幕元素信息、样本触控动作信息,以及样本服务信息。通过对样本屏幕元素信息和样本触控动作信息分别进行编码,得到样本屏幕元素隐信息和样本触控动作隐信息。
在本申请实施例中,对于各个训练样本的样本屏幕元素隐信息、样本触控动作隐信息,通过知识库嵌入方法分别进行解析和编码,得到样本屏幕元素隐信息和样本触控动作隐信息,然后通过协同联合学习,将样本屏幕元素隐信息和样本触控动作隐信息进行融合,得到样本融合信息。
在本申请实施例中,根据多个训练样本的样本融合信息,以及多个样本服务信息中应用的选择结果,得到各个应用被选择的次数(或概率),从而确定各个应用的样本序列行为信息。
在本申请实施例中,根据预设的训练数据,确定各个训练样本的样本融合信息,以及各个应用的样本序列行为信息,以便与用户界面对应的屏幕元素信息和触控操作信息进行协同联合学习和相似度处理,从而得到各个应用的关联度。
在一些实施例中,上述基于屏幕元素信息和触控操作信息,在各个应用上与预设的样本融合信息、预设的各个应用的样本序列行为信息进行关联,从而得到屏幕元素信息与多个应用的各个应用的关联度,可以包括S301-S303。如图3所示,图3为本申请实施例提供的另一种应用推荐方法的可选的步骤流程图。
S301、对屏幕元素信息进行编码,得到屏幕元素隐信息。
在本申请实施例中,可以采用自编码器(auto encoder)对屏幕元素信息进行编码,得到屏幕元素隐信息,自编码器是一种能够通过无监督学习,学到输入数据高效表示的人工神经网络,输入数据的这一高效表示称为编码(codings),其维度一般远小于输入数据,使得自编码器可用于降维。本申请实施例中自编码器可以是任意形式的编码器,对于采用的自编码器的结构不作限制,只要是能够对屏幕元素信息和触控操作信息分别进行编码即可,包括但不限于vanilla自编码器、多层自编码器、卷积自编码器和正则自编码器。其中,正则自编码器包括稀疏自编码器和降噪自编码器,降噪自编码器可以是堆叠降噪自动编码器(stacked denoising auto-encoders,SDAE),卷积自编码器可以是堆叠卷积自编码器(stacked convolutional auto-encoders,SCAE)。
需要说明的是,自编码器通常包括两部分:encoder(也称为识别网络)将输入转换成内部表示,decoder(也称为生成网络)将内部表示转换成输出,输出是在设法重建输入,损失函数是重建损失(reconstruction loss)。在训练过程中,采用自编码器的编码部分(即encoder)对原始图像的信息进行编码,得到高维向量,再采用自编码器的解码部分(即decoder)对高维向量进行解码,得到原始图像。在应用过程中,使用训练完成的自编码器的编码部分(即encoder)对待编码图像进行编码,得到高维向量,高维向量中可以包括带编码图像的全部信息。
S302、对触控操作信息进行编码,得到触控动作隐信息。
在本申请实施例中,可以采用普通的编码方式对触控操作信息进行编码,得到触控动作隐信息。本申请实施例对于编码方式不作限制,只要编码方式能区分开用户进行了哪些触摸操作即可。
示例性的,使用0-1编码方式进行编码,1代表目前用户进行了该项触摸操作,0代表目前用户未进行该项触摸操作。在本申请实施例中,也可以采用其他形式的编码方法对进行和未进行该项触摸操作进行区分,对此本申请实施例不作限制。
S303、基于屏幕元素隐信息、触控动作隐信息、样本融合信息和各个应用的样本序列行为信息进行各个应用的关联,得到各个应用的关联度。
在本申请实施例中,由于屏幕元素信息是与用户界面相关的数据,触控操作信息是与用户当前操作相关的交互数据,因此,采用不同的编码形式对双重模态信息(屏幕元素隐信息和触控动作隐信息)进 行编码,得到的屏幕元素隐信息和触控动作隐信息均是与用户当前操作相关的交互数据,能够满足用户的即时服务需求。基于屏幕元素隐信息、触控动作隐信息、样本融合信息和各个应用的样本序列行为信息进行各个应用的关联,得到各个应用的关联度,提高了各个关联度的准确性。
在一些实施例中,屏幕元素信息包括:用户界面对应的结构型知识;上述S301可以通过以下方式实现。对结构型知识进行编码,得到屏幕元素隐信息。
由于有些用户界面中不存在图像和文本,但是用户界面中一定存在结构型知识,因此本申请实施例对结构型知识进行编码,得到结构信息,将结构信息作为屏幕元素隐信息。示例性的,以编码得到的结构信息是结构向量的表达形式为例进行说明,对于结构型知识,采用关系翻译(Translating Relationship,TransR)方法将结构型知识转化为结构向量,TransR模型能够表征结构型的特征。
在本申请实施例中,对结构型知识进行编码,得到屏幕元素隐信息,提高了屏幕元素隐信息的准确性。
在一些实施例中,屏幕元素信息包括:用户界面对应的结构型知识、文本型知识、视觉型知识和用户界面相关的当前应用服务集合;当前应用服务集合是通过对用户界面进行服务解析确定的。上述图3中的S301可以包括S3011-S3013,如图4所示,图4为本申请实施例提供的再一种应用推荐方法的可选的步骤流程图。
S3011、对结构型知识、文本型知识和视觉型知识分别进行编码,得到结构信息、文本信息和视觉信息。
在本申请实施例中,视觉型知识、文本型知识和结构型知识需要分别通过视觉嵌入、文本嵌入和结构嵌入转换为编码信息。采用数据嵌入方法确定屏幕元素隐信息。示例性的,数据嵌入方法可以是协作知识库嵌入(Collaborative Knowledge base Embedding,CKE)算法,其中,CKE算法包括知识库嵌入(knowledge base embedding)和协同联合学习,其中,知识库嵌入也可以理解为知识图谱嵌入。采用知识库嵌入方法确定屏幕元素隐信息,在后续获得触控动作隐信息之后,可以采用协同联合学习将屏幕元素隐信息和触控动作隐信息进行融合,得到待分析融合信息。以下分别对视觉信息、文本信息和结构信息的编码过程进行说明。其中,对结构型知识进行编码,得到结构信息,具体如上所描述,在此不再赘述。
示例性的,以编码得到的视觉信息是视觉向量的表达形式为例进行说明,对于视觉型知识,可以采用的嵌入方法是堆叠卷积自编码器SCAE,SCAE在训练过程中通过堆叠的卷积神经网络,将原始图像的信息映射(编码)到高维向量,再使用高维向量重建(解码)原始图像,因此,高维向量中包括了原始图像的全部信息。在应用过程中,使用训练完成的SCAE的编码部分对视觉型知识进行编码,得到高维向量。将该高维向量与图像来源分析结果的编码向量进行拼接,得到视觉向量。
示例性的,以编码得到的文本信息是文本向量的表达形式为例进行说明,对于文本型知识,可以采用的嵌入方法是堆叠降噪自动编码器SDAE,SDAE与SCAE的思想类似,在此不再赘述。在应用过程中,使用训练完成的SDAE的编码部分对文本型知识进行编码,得到高维向量,将该高维向量与文本的额外信息的编码向量进行拼接,得到文本向量。
需要说明的是,可以采用任意结构的自编码器对结构型知识、文本型知识进行编码,上述示例仅是以SDAE与SCAE为例进行示例性说明,并不代表本申请实施例局限于此。
在一些实施例中,上述S3011中的结构型知识包括:控件类型、控件功能和控件结构分布的至少一种;文本型知识包括:字符识别信息和特殊符号识别信息中的至少一种,以及文本原始信息;视觉型知识包括:图像内容识别信息和图像来源信息中的至少一种,以及原始图像信息。
在本申请实施例中,控件信息识别得到用户界面上的控件种类、控件功能、控件结构分布等。示例性的,控件种类包括按钮、文本、lable和图像点等,以视频选集为例,控件结构分布包括选集按钮的分布。
在本申请实施例中,文字内容识别得到字符识别信息、链接识别信息和特殊符号识别信息等,并且,对于含有额外信息的文本,示例性的,来自某应用软件的分享链接,会将其额外信息也编码到识别结果中,例如,对于普通网址和购物链接所对应的额外信息是不同的。因此,相对于结构型知识,文本型知识还包括文本原始信息,可以理解为文本本身的信息,不仅能反映字符、特殊符号,还可以反映链接等额外信息。
在本申请实施例中,图像内容识别是图像内容的分析描述(例如,图像本身内容的描述),以及图像内容标签(例如,识别图像中的风景照是哪一个具体的旅游景点,图像是否是表情包等)。图像内容识别还可以包括图像来源分析,图像来源分析会根据图像的样式数据分析图像的来源(例如,分析图像是否来自于拨号界面,分析图像是否是来自某一应用软件界面的截屏)。因此,相对于结构型知识,视觉型知识还包括图像原始信息,可以理解为图像本身的信息,不仅能反映图像内容和图像标签,还可以 反映图像来源等原始信息。
需要说明的是,结构型知识只包括识别得到的结构型数据,也就是说,结构型知识不包括用户界面中与用户交互有关的实质内容,而文本型知识既包括识别得到的字符识别信息和特殊符号识别信息,还包括文本原始信息,视觉型知识既包括识别得到的图像内容识别信息和图像来源信息,还包括原始图像信息。
在本申请实施例中,结构型知识包括控件类型、控件功能和控件结构分布的至少一种;文本型知识包括字符识别信息和特殊符号识别信息中的至少一种,以及文本原始信息;视觉型知识包括图像内容识别信息和图像来源信息中的至少一种,以及原始图像信息,提高了结构型知识、文本型知识、视觉型知识的丰富性。
在一些实施例中,图4中S3011可以包括以下步骤:对结构型知识进行向量化编码,得到结构信息;对字符识别信息和特殊符号识别信息中的至少一种进行向量化编码,得到文本内容理解编码信息;以及对文本原始信息进行堆叠自编码,得到文本高维信息;将文本高维信息和文本内容理解编码信息进行拼接,得到文本信息;对图像内容识别信息和图像来源信息中的至少一种进行信息化编码,得到图像内容理解编码信息;以及对原始图像信息进行堆叠自编码,得到图像高维信息;将图像高维信息和图像内容理解编码信息进行拼接,得到视觉信息。
在本申请实施例中,由于结构型知识包括控件类型、控件功能和控件结构分布的至少一种,没有与用户交互有关的实质内容,因此,在对结构型知识进行编码时,对结构型知识进行向量化编码,即可得到结构信息,提高了结构信息的编码效率。
在本申请实施例中,由于文本型知识包括字符识别信息和特殊符号识别信息中的至少一种,以及文本原始信息,因此,在对文本型知识进行编码时,需要对字符识别信息和特殊符号识别信息中的至少一种进行向量化编码,得到文本内容理解编码信息;并且,对文本原始信息进行堆叠自编码,得到文本高维信息,该文本高维信息能够反映文本原始信息。将文本高维信息和文本内容理解编码信息进行拼接,得到文本信息,提高了文本信息的准确性。
在本申请实施例中,由于视觉型知识包括图像内容识别信息和图像来源信息中的至少一种,以及原始图像信息,因此,在对视觉型知识进行编码时,需要对图像内容识别信息和图像来源信息中的至少一种进行信息化编码,得到图像内容理解编码信息;以及对原始图像信息进行堆叠自编码,得到图像高维信息,该图像高维信息能够反映图像原始信息。将图像高维信息和图像内容理解编码信息进行拼接,得到视觉信息,提高了视觉信息的准确性。
S3012、对当前应用服务集合进行编码,得到服务偏置信息。
在本申请实施例中,由于当前应用需要编码的数据固定,且更新频率低,变化性通常不大,因此,直接简单的将当前应用服务集合进行编码,得到服务偏置信息。
示例性的,由于应用服务集合中包括的应用的数量有多个,并不是所有的应用均与当前用户界面相关,例如,当前用户界面为A聊天界面,在A聊天界面中,对于一个对方发来的购物链接,若用户的触摸操作是“长按”,操作内容是“复制”,则与当前A聊天界面相关的应用服务集合可以包括另一个B应用软件(用于在另一个社交软件中进行链接分享)、C应用软件(用于打开该购物链接,了解该链接中物品的详细信息)、D应用软件(用于在网站中进行查询)。在本申请实施例中,通过对当前用户界面进行服务解析确定当前应用服务集合,可以理解为选择与当前用户界面相关的应用,作为当前应用服务集合,减少需要编码的应用的数量,从而减少资源消耗。由于当前应用服务需要编码的数据固定,且更新频率低,例如,某种浏览器支持打开并解析某一类型的链接,因此,直接简单的对当前应用服务集合进行编码,得到服务偏置信息。采用不同的编号对应用进行区分即可,示例性的,以应用是应用服务、服务偏置信息是向量的表达形式为例进行说明,对应用服务A进行编码得到服务偏置向量0001,对应用服务B进行编码得到服务偏置向量0010。将服务偏置信息和屏幕元素编码信息进行拼接,得到屏幕元素隐信息。
S3013、对文本信息、视觉信息和服务偏置信息中的至少一种,以及结构信息进行拼接,得到屏幕元素隐信息。
在本申请实施例中,由于有些用户界面中不存在图像和文本,即,屏幕元素信息不包括文本信息和视觉信息;有些用户界面中没有与用户界面相关的应用服务集合,即当前应用服务集合为空。可以理解为用户界面中包括结构信息;或者,用户界面中包括结构信息,以及文本信息、视觉信息和服务偏置信息中至少一种。因此,在得到屏幕元素隐信息时,示例性的,可以是对文本信息、视觉信息和服务偏置信息中的至少一种,以及结构信息进行拼接,得到屏幕元素隐信息。也可以是对结构信息、文本信息、视觉信息和服务偏置信息进行拼接,得到屏幕元素隐信息;对此本申请实施例不作限制。其中,在对结构信息、文本信息、视觉信息和服务偏置信息进行拼接,得到屏幕元素隐信息时,若用户界面中不存在 图像、文本,以及对应的当前应用服务集合,则依旧分别对文本型知识、视觉型知识和当前应用服务集合进行编码,编码结果为表征没有实质内容的信息,可以理解为文本信息、视觉信息和服务偏置信息均表征空。由于预设的训练数据中样本融合信息是对结构信息、文本信息、视觉信息和服务偏置信息分别编码后拼接得到的,因此,对待分析数据经过与训练数据一致的处理方法,得到的屏幕元素隐信息,可以保证与样本融合信息的维度一致,在后续计算相似度时,提高计算结果的准确性。
在本申请实施例中,对结构型知识、文本型知识、视觉型知识和当前应用服务集合进行编码分别进行编码,并对编码得到的各个信息进行拼接,得到屏幕元素隐信息,提高了屏幕元素隐信息的准确性。
下面,通过屏幕元素隐向量获取模块,说明本申请实施例在一个实际的应用场景中的示例性应用。以屏幕元素隐信息、触控动作隐信息、服务偏置信息、结构信息、视觉信息、文本信息的表达形式均是向量,以及屏幕元素信息是手机屏幕元素信息为例进行说明。
在本申请实施例中,屏幕元素隐向量获取模块用于获取用户界面的手机屏幕元素的向量表达形式,此处的向量表达形式可以看作是一种基于知识图谱的向量嵌入表达。如图5所示,图5为本申请实施例提供的一种计算屏幕元素隐向量的可选的流程图。图5以应用推荐设备是手机为例进行说明,图5中在计算屏幕元素隐向量时所用到的数据集合,包括当前手机屏幕的用户界面对应的屏幕元素信息,以及当前手机的用户界面所能提供的应用服务集合。
在本申请实施例中,对于手机屏幕元素信息,通过以下三个维度进行解析:控件信息识别、文字内容识别和图像内容识别。图5中以结构型知识表示控件信息识别得到的树状结构型的知识,以文本型知识表示文字内容识别到的信息,以视觉型知识表示图像内容识别到的信息。
需要说明的是,上述手机屏幕元素信息的解析识别,是由用户触控动作中用户操作内容触发的,当用户触控动作满足预设触发条件时,触发上述手机屏幕元素信息的解析识别步骤。
在本申请实施例中,是对控件信息、图像内容和文本内容均进行识别和编码,若用户界面中不存在图像和/或文本,则图像内容和/或文本内容对应的编码向量为表征没有内容的向量,可以理解为编码向量为空,例如,编码向量为0000。在文字内容理解和图像内容理解之前,可以先判断用户界面中是否存在图像和/或文本,若存在图像和/或文本,则对手机屏幕元素信息,进行文字内容理解和/或图像内容理解。
在本申请实施例中,图5中的视觉型知识通过视觉嵌入(embedding)、文本型知识通过文本嵌入(embedding)、结构型知识通过结构嵌入(embedding),转换为对应的编码向量,编码向量包括视觉向量、文本向量和结构向量。并且简单对当前应用服务集合进行编码,得到服务偏置向量,例如,采用不同的编号对应用进行区分即可。将服务偏置向量、视觉向量、文本向量和结构向量进行拼接,得到屏幕元素隐向量。
在一些实施例中,触控操作信息包括:操作内容和触摸操作;在对触控操作信息进行编码,得到触控动作隐信息时,可以通过以下方式实现。对触摸操作和操作内容分别进行编码,得到触摸操作信息和操作内容信息;对触摸操作信息和操作内容信息进行拼接,得到触控动作隐信息。
在本申请实施例中,可以采用普通的编码方式对操作内容和触摸操作进行编码,只要能区分开用户进行了哪些操作,以及哪些手势即可。
示例性的,触摸操作包括但不限于点击、双击、长按、双指捏合、双指放大、滑动和拖拽;使用0-1编码方式进行编码,1代表目前用户做了该项手势,0代表目前用户未做该项手势,在本申请实施例中,也可以采用其他形式的编码方法对是否做了该项手势进行区分。对此本申请实施例不作限制。
示例性的,操作内容包括但不限于复制、转发、分享、拖拽、提醒、保存、收藏和停留等。使用0-1编码方式进行编码,1代表目前用户选择了该项操作,0代表目前用户未选择该项操作,在本申请实施例中,也可以采用其他形式的编码方法对是否选择了该项操作进行区分。对此本申请实施例不作限制。
在本申请实施例中,在对操作内容和触摸操作进行编码,得到操作内容信息和触摸操作信息之后,采用拼接的方式,将操作内容信息和触摸操作信息进行拼接,得到触控动作隐信息,提高了触控动作隐信息的准确性。
下面,通过触控动作隐向量获取模块,说明本申请实施例在一个实际的应用场景中的示例性应用。以触控动作隐信息的表达形式是向量、触控操作信息是用户触控动作为例进行说明。在本申请实施例中,触控动作隐向量获取模块用于获取用户触控动作的向量表达形式,用户触控动作包括用户操作内容和用户触摸操作,如图6所示,图6为本申请实施例提供的一种计算触控动作隐向量的可选的流程图。图6中触控手势表示触摸操作,用户触控手势支持的操作手势包括但不限于点击、双击、长按、双指捏合、双指放大、滑动和拖拽;用户操作内容支持的用户操作包括但不限于复制、转发、分享、拖拽、提醒、保存、收藏和停留等。在对用户操作内容和用户触控手势进行编码,得到用户操作内容向量和用户触控 手势向量之后,采用向量拼接的方式,将用户操作内容向量和用户触控手势向量进行拼接,得到触控动作隐向量。
在一些实施例中,上述图3中S303可以包括S3031-S3033。如图7所示,图7为本申请实施例提供的又一种应用推荐方法的可选的步骤流程图。
S3031、对屏幕元素隐信息和触控动作隐信息进行协同联合学习,得到待分析融合信息。
在本申请实施例中,协同联合学习是将屏幕元素隐信息、触控动作隐信息进行融合,得到待分析融合信息的过程。协同联合学习的主要作用是特征映射和特征降维,采用协同联合学习方法,实现将触控动作隐信息和屏幕元素隐信息作为输入,输出表征用户意图的融合信息,即待分析融合信息。
示例性的,协同联合学习的过程可以理解为采用协同联合模型实现隐信息融合的过程,将屏幕元素隐信息、触控动作隐信息输入训练完成的协同联合模型,输出融合信息,将输出的融合信息作为待分析融合信息,本申请实施例中的协同联合模型可以是任意形式的网络,对于采用的协同联合模型的结构不作限制,只要是能够对屏幕元素隐信息和触控动作隐信息进行融合,得到正确表示融合后的信息即可。包括但不限于映射结构为全卷积网络(fully convolutional network,FCN)。
S3032、获取基于历史行为数据,确定的各个应用的历史序列行为信息。
在本申请实施例中,历史行为数据是用户的历史行为数据,能够反映用户对于各个应用的偏好概率,针对各个应用,结合历史行为数据,得到各个应用的历史序列行为信息。
示例性的,以m个应用为例,针对第k个应用,根据用户的历史行为数据生成历史序列行为信息,记为w k,由此可生成m个历史序列行为信息,历史序列行为信息w k的维度为1×m。
在一些实施例中,图7中S3032可以通过以下方式实现。基于历史行为数据,确定多个应用的历史行为参数;针对每个应用,将该应用对应的历史行为参数,设置为预设参数,结合多个应用中除去该应用之外的历史行为参数,得到该应用的历史序列行为信息。
示例性的,以m个应用、应用对应的历史行为参数是偏好概率、预设参数是预设概率为例。若该用户是新用户,并没有该用户的历史行为数据,也就没有该用户的偏好记录,则该历史序列行为信息中m个应用的偏好概率均相等。若该用户不是新用户,则根据历史行为数据计算m个应用的偏好概率,即,计算历史行为数据中用户选择每个应用的概率。对于m个应用中的第k个应用,将第k个应用的偏好概率设置为预设概率,并结合剩下的m-1个应用的偏好概率,确定第k个应用的历史序列行为信息w k,也就是,针对第k个应用,仅将m个应用中第k个应用的偏好概率设置为预设概率,其他m-1个偏好概率不变,从而形成第k个应用的历史序列行为信息。其中,预设参数可以由本领域技术人员根据实际需求进行适当设置,能够有效对该条应用进行推荐得分的计算即可,以此类推,可以得到m个历史序列行为信息。
在本申请实施例中,针对每个应用,将该应用对应的行为数据,设置为预设参数,结合历史行为数据中除去该应用之外的行为数据,得到该应用的历史序列行为信息,便于后续采用该应用的历史序列行为信息计算相似度,提高相似度的准确性。
S3033、基于各个应用的历史序列行为信息、样本序列行为信息进行相似度处理,以及基于待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到各个应用的相似度。
在本申请实施例中,基于各个应用的历史序列行为信息、样本序列行为信息进行相似度处理,以及基于待分析融合信息和各个训练样本的样本融合信息进行相似度处理。由于训练样本中样本融合信息和样本序列行为信息一一对应,根据以上两次相似度处理,可以得到各个应用的相似度,提高了相似度的准确性。
在一些实施例中,图7中S3033可以包括以下步骤:基于各个应用的历史序列行为信息和样本序列行为信息进行相似度处理,确定各个应用针对各个训练样本的第一相似度信息;将待分析融合信息与各个训练样本的样本融合信息进行相似度处理,得到与各个训练样本对应的第二相似度信息;基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用的相似度。
示例性的,以m个应用为例进行说明,对于m个应用中的第k个应用,在样本序列行为信息中确定各个训练样本的样本序列行为信息,计算该应用的历史序列行为信息,与各个训练样本的样本序列行为信息的相似度,若训练样本的数量为n个,则可以得到n个第一相似度信息。计算待分析融合信息与各个训练样本的样本融合信息的相似度,若训练样本的数量为n个,则可以得到n个第二相似度信息。并基于n个第一相似度信息和n个第二相似度信息,得到第k个应用相似度。重复以上步骤,可以得到m个应用的相似度。
在本申请实施例中,以融合信息和序列行为信息的表达形式均是向量、相似度信息是相似度得分为例进行说明,示例性的,相似度得分可以采用皮尔森相关系数表示。如公式(1)所示:
Figure PCTCN2022120921-appb-000001
示例性的,上述公式(1)中向量x1、x2的相似度表示为sim(x1,x2),皮尔森相关系数定义为两个向量的协方差除以其标准差的乘积。cov(x1,x2)和E[(X1-μx1)(X2-μx2)]表示两个向量的协方差,公式(1)中X1和X2分别表示两个向量,μx1和μx2分别表示两个向量的标量(例如,均值),σ x1和σ x2分别表示两个向量的标准差。
在本申请实施例中,基于历史序列行为信息、样本融合信息分别进行相似度计算,并综合考虑两个维度的相似度信息,得到各个应用的相似度,提高了相似度的准确性。
在一些实施例中,在基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用的相似度时,可以包括以下步骤:基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用针对各个训练样本的第三相似度;针对各个应用,将各个训练样本的第三相似度求和,得到各个应用的相似度。
示例性的,以训练样本的数量为n个为例进行说明,将训练样本的第一相似度信息和该训练样本对应的第二相似度信息相乘,得到第三相似度,由于第一相似度信息和第二相似度信息后的数量均为n个,因此,相乘后可以得到n个第三相似度,一个训练样本对应一个第三相似度,将n个第三相似度求和,得到每个应用的相似度。
在本申请实施例中,以融合信息的表达形式均是向量、相似度信息是相似度得分、历史序列行为信息和样本序列行为信息均是操作序列向量、相似度是推荐得分、训练样本对应各个应用的样本序列行为信息是序列行为矩阵为例进行说明,示例性的,以应用是应用服务为例,待分析数据对应用服务集合中的第k个应用服务的推荐得分可以通过公式(2)进行计算。
score=∑(sim(w k,v i)*sim(s,t i))        (2)
示例性的,上述公式(2)中score表示推荐得分,sim(w k,v i)表示第k个应用服务的操作序列向量w k,与序列行为矩阵中第i条操作序列向量v i的相似度得分,sim(s,t i)表示待分析融合向量s,与融合向量集合中第i条样本融合向量ti的相似度得分,∑表示对n个相似度乘积进行求和。
在本申请实施例中,基于两个维度的相似度信息,得到各个应用的相似度,提高了相似度的准确性。
下面,通过协同联合学习模块和推荐结果生成模块说明本申请实施例在一个实际的应用场景中的示例性应用。以屏幕元素隐信息、触控动作隐信息和融合信息的表达形式均是向量,以及训练样本对应各个应用的样本序列行为信息是序列行为矩阵、历史序列行为信息和样本序列行为信息均是操作序列向量、相似度信息是相似度得分、相似度是推荐得分、各个应用的关联度是应用推荐结果为例进行说明,生成应用推荐结果的流程可以包括:根据手机屏幕元素信息与用户触控动作确定表征用户意图的融合向量,通过协同过滤算法确定得分最高的推荐结果。
在本申请实施例中,协同联合学习模块用于根据触控动作隐向量和屏幕元素隐向量确定待分析融合向量。将触控动作隐向量和屏幕元素隐向量作为输入,输出表征用户意图的融合向量,即待分析融合向量。可以根据推荐结果的反向梯度对协同联合学习模块进行训练。
需要说明的是,本申请实施例中的协同联合学习模块是训练完成的模型,待分析数据以及训练数据均需要进行协同联合学习,以得到融合向量,不同的是,待分析数据经过协同联合学习后得到的是待分析融合向量,训练数据中训练样本经过协同联合学习后得到的是样本融合向量。
在本申请实施例中,应用推荐结果生成模块,用于根据待分析融合向量和预设的训练数据生成应用推荐结果。
示例性的,从根据用户触控动作和手机屏幕元素信息生成应用推荐结果的整体流程的角度,对上述协同联合学习模块和应用推荐结果生成模块分别进行介绍,如图8所示,图8为本申请实施例提供的一种生成推荐结果的可选的流程图。图8中上半部分为训练数据的流程图,训练数据可以表征用户与手机交互产生的数据,图8中下半部分是待分析数据的流程图。以应用是应用服务为例进行说明。
在本申请实施例中,图8中可以通过协同联合学习将样本屏幕元素隐向量和样本触控动作隐向量进行融合,得到表征用户意图的样本融合向量。训练数据包括样本屏幕元素隐向量、样本触控动作隐向量和样本服务信息,样本融合向量与样本服务信息一一对应。示例性的,如果训练数据中包括n条用户交互数据,即n个训练样本,那么在知识库嵌入和协同联合学习进行之后,会得到n条样本融合向量,图8中以融合向量集合包括n条交互数据的样本融合向量。n条交互数据对应的样本服务信息的数量为n,该n个样本服务信息中会存在相同的应用服务。将应用推荐设备所支持的所有应用服务的集合作为应用服务集,其包括的应用服务总数量为m,该m个应用服务中各个应用服务不相同,m远小于n。
需要说明的是,对于训练数据中每个训练样本的样本触控动作信息和样本屏幕元素信息,通过知识库嵌入方法进行解析和编码,得到屏幕元素隐向量和触控动作隐向量,图8中仅示出了训练数据中每个训练样本的样本屏幕元素隐向量和样本触控动作隐向量。
在本申请实施例中,训练数据中包括n个训练样本,一个训练样本包括一条样本融合向量和其对应的样本服务信息。根据n条样本融合向量和其对应的n个样本服务信息,确定各个应用服务的样本序列行为信息,即,图8中右上角的维度为n×m的序列行为矩阵。图8中的Rij表示第i条样本融合向量对第j个应用服务的偏好概率,也可以理解为第i条交互数据对第j个应用服务的交互得分。交互得分由训练数据中样本融合向量对应的样本服务信息的数量确定,也可以理解为在同一交互数据下,用户对各个应用服务的偏好概率(即选择频率),偏好概率越高,Rij值越大。
示例性的,假设只存在两种应用服务,即,j=1或2,第i条交互数据选择了第一种应用服务3次,第二种应用服务1次,则Ri1为3/(1+3)=0.75,相应的,Ri2为0.25。对融合向量集合中的所有交互数据进行上述计算,可以得到维度为n×m的序列行为矩阵。将序列行为矩阵中的每一行的行向量作为一条操作序列向量,可以得到n条操作序列向量,第i条操作序列向量记为v i,图8中v n表示第n条操作序列向量。
在本申请实施例中,对于待分析数据,获取触控动作信息和屏幕元素信息,通过知识库嵌入方法进行解析和编码,得到屏幕元素隐向量和触控动作隐向量对屏幕元素隐向量和触控动作隐向量进行协同联合学习,得到待分析融合向量,记为s。其中,待分析融合向量s与第i条样本融合向量的维度相同。
在本申请实施例中,对于待分析数据,在图8的组合服务模块中,针对m个应用服务中的第k个应用服务,根据用户的历史行为数据生成待分析数据的操作序列向量,记为w k,由此可生成m个操作序列向量。操作序列向量w k的维度与上述序列行为矩阵中每一行的行向量的维度一致,也就是与第i条操作序列向量v i的维度一致,即1×m。在组合服务模块生成待分析数据的操作序列向量时,示例性的,对于m个应用服务中的第k个应用服务,根据用户的历史行为数据计算m个应用服务的偏好概率,将第k个应用服务的偏好概率设置为预设峰值(即预设参数),并结合剩下的m-1个应用服务偏好概率,确定第k个应用服务的操作序列向量w k。以此类推,可以得到m个操作序列向量。
在本申请实施例中,推荐结果生成模块使用协同过滤算法可以生成最可能的应用推荐。示例性的,对于m个应用服务中的第k个应用服务,需要计算待分析数据的操作序列向量w k,与融合向量集合中的每条样本融合向量对应的操作序列向量v i的相似度得分,n条样本融合向量对应的n个操作序列向量,因此可以得到n个相似度得分。还需要计算待分析融合向量s与融合向量集合中每条样本融合向量(记为t i)的相似度得分,融合向量集合中包括n条样本融合向量,因此可以得到n个相似度得分。
在本申请实施例中,将上述n个操作序列向量的相似度得分和n个融合向量的相似度得分,对应的相乘后求和,得到第k个应用服务的推荐得分。重复以上步骤,计算待分析数据对应用服务集合中的m个应用服务的推荐得分,可以得到m个应用服务的推荐得分。选取推荐得分最高的应用服务,得到应用推荐结果,实现了结合手机屏幕元素信息和用户触控动作得到应用推荐结果。
本申请实施例提出了一种结合手机屏幕元素信息与用户触控动作的应用推荐方法,综合考虑用户的具体触控动作和用户正在交互的屏幕内容对应的元素信息。通过提取屏幕元素信息与用户触控动作所代表的意图隐向量,并将其应用到基于协同过滤算法的智能应用推荐中,推测用户的强交互需求,再进行相应的应用推荐,能够满足用户即时需求。且结合触控模态信息以及用户界面的屏幕元素信息,采用与用户强相关的双重模态信息向用户进行智能推荐服务,提高了应用推荐结果的准确性。
在本申请实施例中,结合智能交互场景中用户触控动作与相应的屏幕元素信息中蕴含的用户主动强需求,充分考虑了用户的主动意图,相较于相关技术中由用户被动接收智能推荐服务,提供了一种用户意图主动触发的智能推荐服务,提高了推荐结果的准确性。通过识别用户意图,向用户提供个性化、定制化的服务,提高了用户与设备之间的交互效率。
在一些实施例中,上述图2中S204还可以通过以下方式实现。基于满足预设条件的应用,生成服务提示信息;在用户界面上,推荐显示服务提示信息。
在本申请实施例中,基于满足预设条件的应用,生成服务提示信息,服务提示信息可以是不同类型的应用软件列表,在用户界面推荐显示服务提示信息,以供用户进行选择。服务提示信息可以以悬浮窗或按钮形式显示在用户界面中的任意位置,本申请实施例对于服务提示信息的形式和位置不作限制。
示例性的,如图9和图10所示,图9和图10为本申请实施例提供了一种应用推荐结果的示例性的表现形式示意图。图9和图10中的智能服务推荐悬浮窗是服务提示信息的一种表现形式。在图9中,识别到当前用户界面(User Interface,UI)显示用户接收到其他用户(刘总)发送来的某购物应用程序的分享链接,同时检测到用户的交互强意图:用户长按了该分享链接,并点击复制,本申请实施例的应用推荐方法能够准确分析到用户的潜在交互意图:用户想要在该购物应用程序中打开该分享链接。因此, 应用推荐设备作出推荐服务,并在手机屏幕下方通过悬浮窗的形式显示应用推荐结果,用户点击智能服务推荐悬浮窗的控件1,即可跳转到对应购物应用软件A中打开链接。
示例性的,在图10中,应用推荐设备识别到当前用户界面接收到一个来自王哥的社交会话提供的某视频网站的截图分享,同时检测到用户的交互强意图:用户长按了该链接并点击了查看原图。本申请实施例的应用推荐方法能够识别到用户的潜在意图:在该视频应用中打开该视频。因此,应用推荐设备做出推荐服务,并在手机屏幕下方通过悬浮窗的形式显示应用推荐结果,用户点击智能服务推荐悬浮窗的控件2,即可跳转到该视频应用的对应视频界面,查看视频。
在申请实施例中,还基于满足预设条件的应用,生成服务提示信息;在用户界面上,推荐显示服务提示信息,实现了快速有效的转发用户在用户界面中选中的数据。
本申请实施例还提供一种应用推荐方法,如图11所示,图11为本申请实施例提供的又一种应用推荐方法的可选的步骤流程图,应用推荐方法包括以下步骤:
S1101、获取针对用户界面的触控操作信息,以及获取用户界面对应的屏幕元素信息。
S1102、基于屏幕元素信息和触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取屏幕元素信息与多个应用的关联度。
S1103、对关联度满足预设条件的应用进行推荐显示。
在本申请实施例中,上述多个应用可以为同一类型应用,也可以为不同类型的应用,对此本申请实施例不作限制。
在一些实施例中,上述S1102还可以包括以下步骤:对屏幕元素信息进行编码,得到屏幕元素隐信息;对触控操作信息进行编码,得到触控动作隐信息;对屏幕元素隐信息和触控动作隐信息进行协同联合学习,得到待分析融合信息;获取基于历史行为数据,确定的各个应用的历史序列行为信息;基于各个应用的历史序列行为信息、样本序列行为信息进行相似度处理,以及基于待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到各个应用的相似度。
在一些实施例中,上述在基于各个应用的历史序列行为信息、样本序列行为信息进行相似度处理,以及基于待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到各个应用的相似度时,可以通过以下方式实现。基于各个应用的历史序列行为信息和样本序列行为信息进行相似度处理,确定各个应用针对各个训练样本的第一相似度信息;将待分析融合信息与各个训练样本的样本融合信息进行相似度处理,得到与各个训练样本对应的第二相似度信息;基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用的相似度。
为实现本申请实施例的应用推荐方法,本申请实施例还提供一种应用推荐装置,如图12所示,图12为本申请实施例提供的一种应用推荐装置的结构示意图,该应用推荐装置120包括:第一获取部分1201,被配置为获取作用于用户界面的触控操作信息;以及基于触控操作信息,获取用户界面对应的屏幕元素信息;第一关联部分1202,被配置为基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;第一显示部分1203,被配置为对关联度满足预设条件的应用进行推荐显示。
在一些实施例中,触控操作信息包括:触摸操作和操作内容;
第一获取部分1201,还被配置为获取作用于用户界面的触摸操作;显示与触摸操作对应的操作内容。
在一些实施例中,触控操作信息包括:触摸操作和操作内容,第一获取部分1201,还被配置为获取作用于用户界面的触摸操作;基于触摸操作,获取用户界面对应的屏幕元素信息;基于触摸操作和屏幕元素信息,显示对应的操作内容。
在一些实施例中,第一关联部分1202,还被配置为响应于对操作内容的选择,获得用户界面对应的屏幕元素信息与多个应用的关联度。
在一些实施例中,触控操作信息包括:触摸操作和操作内容;
第一获取部分1201,还被配置为在针对用户界面的选择操作的作用下,显示被选中的内容对应的操作内容;在预设时间范围内,获取针对操作内容的触摸操作;触摸操作包括选择操作。
在一些实施例中,第一关联部分1202,还被配置为基于屏幕元素信息和触控操作信息,在各个应用上与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联,从而得到屏幕元素信息与多个应用的各个应用的关联度。
在一些实施例中,第一关联部分1202包括第一屏幕元素隐信息编码部分、第一触控动作隐信息编码部分和第一关联度处理部分;
第一屏幕元素隐信息编码部分,被配置为对屏幕元素信息进行编码,得到屏幕元素隐信息;
第一触控动作隐信息编码部分,被配置为对触控操作信息进行编码,得到触控动作隐信息;
第一关联度处理部分,被配置为基于屏幕元素隐信息、触控动作隐信息、样本融合信息和各个应用的样本序列行为信息进行各个应用的关联,得到各个应用的关联度。
在一些实施例中,屏幕元素信息包括:用户界面对应的结构型知识;
第一屏幕元素隐信息编码部分,还被配置为对结构型知识进行编码,得到屏幕元素隐信息。
在一些实施例中,屏幕元素信息包括:用户界面对应的结构型知识、文本型知识、视觉型知识和用户界面相关的当前应用服务集合;当前应用服务集合是通过对用户界面进行服务解析确定的;
第一屏幕元素隐信息编码部分,还被配置为对结构型知识、文本型知识和视觉型知识分别进行编码,得到结构信息、文本信息和视觉信息;对当前应用服务集合进行编码,得到服务偏置信息;对文本信息、视觉信息和服务偏置信息中的至少一种,以及结构信息进行拼接,得到屏幕元素隐信息。
在一些实施例中,结构型知识包括:控件类型、控件功能和控件结构分布的至少一种;文本型知识包括:字符识别信息和特殊符号识别信息中的至少一种,以及文本原始信息;视觉型知识包括:图像内容识别信息和图像来源信息中的至少一种,以及原始图像信息。
在一些实施例中,第一屏幕元素隐信息编码部分,还被配置为对结构型知识进行向量化编码,得到结构信息;对字符识别信息和特殊符号识别信息中的至少一种进行向量化编码,得到文本内容理解编码信息;以及对文本原始信息进行堆叠自编码,得到文本高维信息;将文本高维信息和文本内容理解编码信息进行拼接,得到文本信息;对图像内容识别信息和图像来源信息中的至少一种进行信息化编码,得到图像内容理解编码信息;以及对原始图像信息进行堆叠自编码,得到图像高维信息;将图像高维信息和图像内容理解编码信息进行拼接,得到视觉信息。
在一些实施例中,触控操作信息包括:操作内容和触摸操作;
第一触控动作隐信息编码部分,还被配置为对触摸操作和操作内容分别进行编码,得到触摸操作信息和操作内容信息;对触摸操作信息和操作内容信息进行拼接,得到触控动作隐信息。
在一些实施例中,第一获取部分1201,还被配置为对用户界面或待识别区域进行控件信息识别,得到结构型知识;对用户界面或待识别区域进行文字内容识别,得到文本型知识;对用户界面或待识别区域进行图像内容识别,得到视觉型知识;对用户界面进行服务解析,确定当前应用服务集合;文本型知识、视觉型知识和当前应用服务集合中的至少一种,以及结构型知识作为屏幕元素信息。
在一些实施例中,第一获取部分1201,还被配置为通过无障碍服务接口,获取用户界面或待识别区域对应的控件树的结构信息;对控件树的结构信息进行控件信息识别,得到结构型知识。
在一些实施例中,第一关联部分1202还包括第一协同联合学习部分(对应于上述协同联合学习模块);
第一协同联合学习部分,被配置为对屏幕元素隐信息和触控动作隐信息进行协同联合学习,得到待分析融合信息;
第一获取部分1201,还被配置为获取基于历史行为数据,确定的各个应用的历史序列行为信息;
第一关联度处理部分,还被配置为基于各个应用的历史序列行为信息、样本序列行为信息进行相似度处理,以及基于待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到各个应用的相似度。
在一些实施例中,第一关联度处理部分,还被配置为基于历史行为数据,确定多个应用的历史行为参数;针对每个应用,将该应用对应的历史行为参数,设置为预设参数,结合多个应用中除去该应用之外的历史行为参数,得到该应用的历史序列行为信息。
在一些实施例中,第一关联度处理部分,还被配置为基于各个应用的历史序列行为信息和样本序列行为信息进行相似度处理,确定各个应用针对各个训练样本的第一相似度信息;将待分析融合信息与各个训练样本的样本融合信息进行相似度处理,得到与各个训练样本对应的第二相似度信息;基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用的相似度。
在一些实施例中,第一关联度处理部分,还被配置为基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用针对各个训练样本的第三相似度;针对各个应用,将各个训练样本的第三相似度求和,得到各个应用的相似度。
在一些实施例中,应用推荐装置120还包括训练部分,训练部分,被配置为根据预设的训练数据,确定各个训练样本对应的样本屏幕元素隐信息、样本触控动作隐信息和样本服务信息;对样本屏幕元素隐信息和样本触控动作隐信息进行协同联合学习,得到各个训练样本的预设的样本融合信息;根据各个训练样本的样本服务信息,确定预设的各个应用的样本序列行为信息。
在一些实施例中,第一显示部分1203,还被配置为在用户界面,推荐显示各个应用的相似度最高的应用;或者,在用户界面,推荐显示各个应用的相似度最高的预设数量的应用。
在一些实施例中,应用包括:应用类型,和/或,应用服务。
在一些实施例中,第一显示部分1203,还被配置为基于满足预设条件的应用,生成服务提示信息;在用户界面上,推荐显示服务提示信息。
在一些实施例中,第一获取部分1201,还被配置为在满足触控操作信息表征其操作内容满足预设触发推荐条件的情况下,获取用户界面对应的全界面的屏幕元素信息;预设触发推荐条件表征预期的操作意图。
在一些实施例中,第一获取部分1201,还被配置为在触控操作信息作用的触控位置,确定当前待识别区域;对当前待识别区域进行识别,确定屏幕元素信息。
在一些实施例中,应用推荐装置120包括识别部分,识别部分,被配置为在满足当前待识别区域中未识别出屏幕元素信息的情况下,确定下一个待识别区域,在下一个待识别区域进行识别,直至得到存在屏幕元素信息或者识别完全界面时为止;下一个待识别区域大于当前待识别区域。
为实现本申请实施例的应用推荐方法,本申请实施例还提供另一种应用推荐装置,如图13所示,图13为本申请实施例提供的一种应用推荐装置的结构示意图,该应用推荐装置130包括:第二获取部分1301,被配置为获取针对用户界面的触控操作信息,以及获取用户界面对应的屏幕元素信息;第二关联部分1302,被配置为基于屏幕元素信息和触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取屏幕元素信息与多个应用的关联度;第二显示部分1303,被配置为对关联度满足预设条件的应用进行推荐显示。
在一些实施例中,第二关联部分1302包括第二屏幕元素隐信息编码部分、第二触控动作隐信息编码部分、第二协同联合学习部分(对应于上述协同联合学习模块)和第二关联度处理部分;
第二屏幕元素隐信息编码部分,被配置为对屏幕元素信息进行编码,得到屏幕元素隐信息;
第二触控动作隐信息编码部分,被配置为对触控操作信息进行编码,得到触控动作隐信息;
第二协同联合学习部分,被配置为对屏幕元素隐信息和触控动作隐信息进行协同联合学习,得到待分析融合信息;
第二获取部分1301,还被配置为获取基于历史行为数据,确定的各个应用的历史序列行为信息;
第二关联度处理部分,被配置为基于各个应用的历史序列行为信息、样本序列行为信息进行相似度处理,以及基于待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到各个应用的相似度。
在一些实施例中,第二关联度处理部分,被配置为基于各个应用的历史序列行为信息和样本序列行为信息进行相似度处理,确定各个应用针对各个训练样本的第一相似度信息;将待分析融合信息与各个训练样本的样本融合信息进行相似度处理,得到与各个训练样本对应的第二相似度信息;基于各个应用针对各个训练样本的第一相似度信息,以及各个训练样本对应的第二相似度信息,得到各个应用的相似度。
需要说明的是,上述实施例提供的应用推荐装置在进行应用推荐时,仅以上述各程序部分的划分进行举例说明,实际应用中,可以根据需要而将上述处理分配由不同的程序部分完成,即将装置的内部结构划分成不同的程序部分,以完成以上描述的全部或者部分处理。另外,上述实施例提供的应用推荐装置与应用推荐方法实施例属于同一构思,其具体实现过程及有益效果详见方法实施例,这里不再赘述。对于本装置实施例中未披露的技术细节,请参照本申请方法实施例的描述而理解。
在本申请实施例中,图14为本申请实施例提出的应用推荐设备组成结构示意图,如图14所示,本申请实施例提出的应用推荐设备140还可以包括处理器1401和存储器1402,存储器1402存储有可在处理器1401上运行的计算机程序,在一些实施例中,应用推荐设备140还可以包括通信接口1403和总线1404,总线1404被配置为连接处理器1401、存储器1402以及通信接口1403。
在本申请实施例中,总线1404,被配置为连接通信接口1403、处理器1401以及存储器1402以及这些器件之间的相互通信。
在本申请实施例中,上述处理器1401,被配置为获取作用于用户界面的触控操作信息;基于触控操作信息,获取用户界面对应的屏幕元素信息;基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;对关联度满足预设条件的应用进行推荐显示。
在本申请实施例中,上述处理器1401,还被配置为获取针对用户界面的触控操作信息,以及获取用户界面对应的屏幕元素信息;基于屏幕元素信息和触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;对关联度满足预设条件的应用进行推荐显示。
在本申请实施例中,上述处理器1401可以为特定用途集成电路(Application Specific Integrated Circuit,ASIC)、数字信号处理器(Digital Signal Processor,DSP)、数字信号处理装置(Digital Signal  Processing Device,DSPD)、可编程逻辑装置(ProgRAMmable Logic Device,PLD)、现场可编程门阵列(Field ProgRAMmable Gate Array,FPGA)、中央处理器(Central Processing Unit,CPU)、控制器、微控制器、微处理器中的至少一种。可以理解地,对于不同的设备,用于实现上述处理器功能的电子器件还可以为其它,本申请实施例不作具体限定。
应用推荐设备140中存储器1402可以与处理器1401连接,存储器1402被配置为存储可执行程序代码和数据,该程序代码包括计算机操作指令。存储器1402均可能包括高速RAM存储器,也可能还包括非易失性存储器,例如,至少两个磁盘存储器。在实际应用中,上述存储器1402均可以是易失性存储器(volatile memory),例如随机存取存储器(Random-Access Memory,RAM);或者非易失性存储器(non-volatile memory),例如只读存储器(Read-Only Memory,ROM),快闪存储器(flash memory),硬盘(Hard Disk Drive,HDD)或固态硬盘(Solid-State Drive,SSD);或者上述种类的存储器的组合,并向处理器1401提供指令和数据。
另外,在本实施例中的各功能部分可以集成在一个处理部分中,也可以是各个部分单独物理存在,也可以两个或两个以上部分集成在一个部分中。上述集成的部分既可以采用硬件的形式实现,也可以采用软件功能部分的形式实现。
集成的部分如果以软件功能部分的形式实现并非作为独立的产品进行销售或使用时,可以存储在一个计算机可读取存储介质中,基于这样的理解,本实施例的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来,该计算机软件产品存储在一个存储介质中,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)或processor(处理器)执行本实施例方法的全部或部分步骤。而前述的存储介质包括:U盘、移动硬盘、只读存储器(Read Only Memory,ROM)、随机存取存储器(Random Access Memory,RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
本申请实施例提供一种计算机可读存储介质,其上存储有程序,该程序被处理器执行时实现如上任一实施例的应用推荐方法。示例性的,本实施例中的一种应用推荐方法对应的程序指令可以被存储在光盘,硬盘,U盘等存储介质上,当存储介质中的与一种应用推荐方法对应的程序指令被一电子设备读取或被执行时,可以实现如上述任一实施例的应用推荐方法。
本领域内的技术人员应明白,本申请实施例可提供为方法、系统、或计算机程序产品。因此,本申请可采用硬件实施例、软件实施例、或结合软件和硬件方面的实施例的形式。而且,本申请可采用在一个或多个其中包括有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器和光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的实现流程示意图和/或方框图来描述的。应理解可由计算机程序指令实现流程示意图和/或方框图中的每一流程和/或方框、以及实现流程示意图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器,使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在实现流程示意图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中,使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品,该指令装置实现在实现流程示意图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上,使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理,从而在计算机或其他可编程设备上执行的指令提供用于实现在实现流程示意图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。以上,为本申请的较佳实施例而已,并非用于限定本申请的保护范围。
工业实用性
本申请实施例公开了一种应用推荐方法、装置、设备及计算机可读存储介质。该方法包括:获取作用于用户界面的触控操作信息;基于触控操作信息,获取用户界面对应的屏幕元素信息;屏幕元素信息和触控操作信息均是与用户当前操作相关的交互数据,能够满足用户的即时服务需求。基于触控操作信息和屏幕元素信息,获取屏幕元素信息与多个应用的关联度,多个应用为同一类型应用;对关联度满足预设条件的应用进行推荐显示。本申请实施例通过结合屏幕元素信息和触控操作信息的双重模态信息,从两个维度对用户界面的交互数据进行考虑,然后结合与多个应用的关联度,提高了应用推荐结果的准确性。

Claims (32)

  1. 一种应用推荐方法,所述方法包括:
    获取作用于用户界面的触控操作信息;
    基于所述触控操作信息,获取所述用户界面对应的屏幕元素信息;
    基于所述触控操作信息和所述屏幕元素信息,获取所述屏幕元素信息与多个应用的关联度,所述多个应用为同一类型应用;
    对所述关联度满足预设条件的应用进行推荐显示。
  2. 根据权利要求1所述的方法,其中,所述触控操作信息包括:触摸操作和操作内容;所述获取作用于用户界面的触控操作信息,包括:
    获取作用于用户界面的所述触摸操作;
    显示与所述触摸操作对应的所述操作内容。
  3. 根据权利要求1所述的方法,其中,所述触控操作信息包括:触摸操作和操作内容;所述获取作用于用户界面的触控操作信息,包括:
    获取作用于用户界面的所述触摸操作;
    基于所述触摸操作,获取所述用户界面对应的所述屏幕元素信息;
    基于所述触摸操作和所述屏幕元素信息,显示对应的所述操作内容。
  4. 根据权利要求2或3所述的方法,其中,所述基于所述触控操作信息和所述屏幕元素信息,获取所述屏幕元素信息与多个应用的关联度,包括:
    响应于对所述操作内容的选择,获得所述用户界面对应的所述屏幕元素信息与所述多个应用的关联度。
  5. 根据权利要求1所述的方法,其中,所述触控操作信息包括:触摸操作和操作内容;所述获取作用于用户界面的触控操作信息,包括:
    在针对所述用户界面的选择操作的作用下,显示被选中的内容对应的所述操作内容;
    在预设时间范围内,获取针对所述操作内容的所述触摸操作;所述触摸操作包括所述选择操作。
  6. 根据权利要求1至5任一项所述的方法,其中,所述基于所述触控操作信息和所述屏幕元素信息,获取所述屏幕元素信息与多个应用的关联度,包括:
    基于所述屏幕元素信息和所述触控操作信息,在各个应用上与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联,从而得到所述屏幕元素信息与所述多个应用的各个应用的所述关联度。
  7. 根据权利要求6所述的方法,其中,所述基于所述屏幕元素信息和所述触控操作信息,在各个应用上与预设的样本融合信息、预设的各个应用的样本序列行为信息进行关联,从而得到所述屏幕元素信息与所述多个应用的各个应用的关联度,包括:
    对所述屏幕元素信息进行编码,得到屏幕元素隐信息;
    对所述触控操作信息进行编码,得到触控动作隐信息;
    基于所述屏幕元素隐信息、所述触控动作隐信息、所述样本融合信息和各个应用的所述样本序列行为信息进行各个应用的关联,得到所述各个应用的关联度。
  8. 根据权利要求7所述的方法,其中,所述屏幕元素信息包括:用户界面对应的结构型知识;
    所述对所述屏幕元素信息进行编码,得到屏幕元素隐信息,包括:
    对所述结构型知识进行编码,得到所述屏幕元素隐信息。
  9. 根据权利要求7所述的方法,其中,所述屏幕元素信息包括:用户界面对应的结构型知识、文本型知识、视觉型知识和用户界面相关的当前应用服务集合;所述当前应用服务集合是通过对所述用户界面进行服务解析确定的;
    所述对所述屏幕元素信息进行编码,得到屏幕元素隐信息,包括:
    对所述结构型知识、所述文本型知识和所述视觉型知识分别进行编码,得到结构信息、文本信息和视觉信息;
    对所述当前应用服务集合进行编码,得到服务偏置信息;
    对所述文本信息、所述视觉信息和所述服务偏置信息中的至少一种,以及所述结构信息进行拼接,得到所述屏幕元素隐信息。
  10. 根据权利要求9所述的方法,其中,
    所述结构型知识包括:控件类型、控件功能和控件结构分布的至少一种;
    所述文本型知识包括:字符识别信息和特殊符号识别信息中的至少一种,以及文本原始信息;
    所述视觉型知识包括:图像内容识别信息和图像来源信息中的至少一种,以及原始图像信息。
  11. 根据权利要求9或10所述的方法,其中,所述对所述结构型知识、所述文本型知识和所述视觉型知识分别进行编码,得到结构信息、文本信息和视觉信息,包括:
    对所述结构型知识进行向量化编码,得到所述结构信息;
    对字符识别信息和特殊符号识别信息中的至少一种进行向量化编码,得到文本内容理解编码信息;以及对文本原始信息进行堆叠自编码,得到文本高维信息;
    将所述文本高维信息和所述文本内容理解编码信息进行拼接,得到所述文本信息;
    对图像内容识别信息和图像来源信息中的至少一种进行信息化编码,得到图像内容理解编码信息;以及对原始图像信息进行堆叠自编码,得到图像高维信息;
    将所述图像高维信息和所述图像内容理解编码信息进行拼接,得到所述视觉信息。
  12. 根据权利要求7所述的方法,其中,所述触控操作信息包括:操作内容和触摸操作;
    所述对所述触控操作信息进行编码,得到触控动作隐信息,包括:
    对所述触摸操作和所述操作内容分别进行编码,得到触摸操作信息和操作内容信息;
    对所述触摸操作信息和所述操作内容信息进行拼接,得到所述触控动作隐信息。
  13. 根据权利要求1至12任一项所述的方法,其中,所述获取所述用户界面对应的屏幕元素信息,包括:
    对所述用户界面或待识别区域进行控件信息识别,得到结构型知识;
    对所述用户界面或待识别区域进行文字内容识别,得到文本型知识;
    对所述用户界面或待识别区域进行图像内容识别,得到视觉型知识;
    对所述用户界面进行服务解析,确定当前应用服务集合;
    所述文本型知识、所述视觉型知识和所述当前应用服务集合中的至少一种,以及所述结构型知识作为所述屏幕元素信息。
  14. 根据权利要求13所述的方法,其中,所述对所述用户界面或待识别区域进行控件信息识别,得到结构型知识,包括:
    通过无障碍服务接口,获取所述用户界面或所述待识别区域对应的控件树的结构信息;
    对所述控件树的结构信息进行控件信息识别,得到所述结构型知识。
  15. 根据权利要求7所述的方法,其中,所述基于所述屏幕元素隐信息、所述触控动作隐信息、所述样本融合信息和各个应用的所述样本序列行为信息进行各个应用的关联,得到所述各个应用的关联度,包括:
    对所述屏幕元素隐信息和所述触控动作隐信息进行协同联合学习,得到待分析融合信息;
    获取基于历史行为数据,确定的各个应用的历史序列行为信息;
    基于各个应用的所述历史序列行为信息、所述样本序列行为信息进行相似度处理,以及基于所述待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到所述各个应用的相似度。
  16. 根据权利要求15所述的方法,其中,所述获取基于历史行为数据,确定的各个应用的历史序列行为信息,包括:
    基于所述历史行为数据,确定多个应用的历史行为参数;
    针对每个应用,将该应用对应的历史行为参数,设置为预设参数,结合所述多个应用中除去该应用之外的历史行为参数,得到该应用的历史序列行为信息。
  17. 根据权利要求15所述的方法,其中,所述基于各个应用的所述历史序列行为信息、所述样本序列行为信息进行相似度处理,以及基于所述待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到所述各个应用的相似度,包括:
    基于各个应用的所述历史序列行为信息和所述样本序列行为信息进行相似度处理,确定各个应用针对各个训练样本的第一相似度信息;
    将所述待分析融合信息与所述各个训练样本的样本融合信息进行相似度处理,得到与所述各个训练样本对应的第二相似度信息;
    基于各个应用针对各个训练样本的第一相似度信息,以及所述各个训练样本对应的第二相似度信息,得到所述各个应用的相似度。
  18. 根据权利要求17所述的方法,其中,所述基于各个应用针对各个训练样本的第一相似度信息,以及所述各个训练样本对应的第二相似度信息,得到所述各个应用的相似度,包括:
    基于各个应用针对各个训练样本的第一相似度信息,以及所述各个训练样本对应的第二相似度信息,得到各个应用针对各个训练样本的第三相似度;
    针对各个应用,将各个训练样本的第三相似度求和,得到所述各个应用的相似度。
  19. 根据权利要求6所述的方法,其中,所述基于所述屏幕元素信息和所述触控操作信息,在各个应用上与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联,从而得到所述屏幕元素信息与所述多个应用的各个应用的所述关联度之前,所述方法还包括:
    根据预设的训练数据,确定各个训练样本对应的样本屏幕元素隐信息、样本触控动作隐信息和样本服务信息;
    对所述样本屏幕元素隐信息和所述样本触控动作隐信息进行协同联合学习,得到所述各个训练样本的预设的所述样本融合信息;
    根据各个训练样本的所述样本服务信息,确定预设的各个应用的所述样本序列行为信息。
  20. 根据权利要求1至19任一项所述的方法,其中,所述对所述关联度满足预设条件的应用进行推荐显示,包括:
    在所述用户界面,推荐显示所述各个应用的相似度最高的应用;
    或者,
    在所述用户界面,推荐显示所述各个应用的相似度最高的预设数量的应用。
  21. 根据权利要求1至19任一项所述的方法,其中,
    应用包括:应用类型,和/或,应用服务。
  22. 根据权利要求1至19任一项所述的方法,其中,所述对所述关联度满足预设条件的应用进行推荐显示,包括:
    基于所述满足预设条件的应用,生成服务提示信息;
    在所述用户界面上,推荐显示所述服务提示信息。
  23. 根据权利要求1至19任一项所述的方法,其中,所述基于所述触控操作信息,获取所述用户界面对应的屏幕元素信息,包括:
    在满足所述触控操作信息表征其操作内容满足预设触发推荐条件的情况下,获取所述用户界面对应的全界面的屏幕元素信息;所述预设触发推荐条件表征预期的操作意图。
  24. 根据权利要求1至19任一项所述的方法,其中,所述基于所述触控操作信息,获取所述用户界面对应的屏幕元素信息,包括:
    在所述触控操作信息作用的触控位置,确定当前待识别区域;
    对所述当前待识别区域进行识别,确定所述屏幕元素信息。
  25. 根据权利要求24所述的方法,其中,所述在所述触控操作信息作用的触控位置,确定当前待识别区域之后,所述方法还包括:
    在满足所述当前待识别区域中未识别出屏幕元素信息的情况下,确定下一个待识别区域,在所述下一个待识别区域进行识别,直至得到存在所述屏幕元素信息或者识别完全界面时为止;所述下一个待识别区域大于所述当前待识别区域。
  26. 一种应用推荐方法,所述方法包括:
    获取针对用户界面的触控操作信息,以及获取所述用户界面对应的屏幕元素信息;
    基于所述屏幕元素信息和所述触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取所述屏幕元素信息与多个应用的关联度;
    对所述关联度满足预设条件的应用进行推荐显示。
  27. 根据权利要求26所述的方法,其中,所述基于所述屏幕元素信息和所述触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取所述屏幕元素信息与多个应用的关联度,包括:
    对所述屏幕元素信息进行编码,得到屏幕元素隐信息;
    对所述触控操作信息进行编码,得到触控动作隐信息;
    对所述屏幕元素隐信息和所述触控动作隐信息进行协同联合学习,得到待分析融合信息;
    获取基于历史行为数据,确定的各个应用的历史序列行为信息;
    基于各个应用的所述历史序列行为信息、所述样本序列行为信息进行相似度处理,以及基于所述待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到所述各个应用的相似度。
  28. 根据权利要求27所述的方法,其中,所述基于各个应用的所述历史序列行为信息、所述样本序列行为信息进行相似度处理,以及基于所述待分析融合信息和各个训练样本的样本融合信息进行相似度处理,得到所述各个应用的相似度,包括:
    基于各个应用的所述历史序列行为信息和所述样本序列行为信息进行相似度处理,确定各个应用针对各个训练样本的第一相似度信息;
    将所述待分析融合信息与所述各个训练样本的样本融合信息进行相似度处理,得到与所述各个训练 样本对应的第二相似度信息;
    基于各个应用针对各个训练样本的第一相似度信息,以及所述各个训练样本对应的第二相似度信息,得到所述各个应用的相似度。
  29. 一种应用推荐装置,所述装置包括:
    第一获取部分,被配置为获取作用于用户界面的触控操作信息;以及基于所述触控操作信息,获取所述用户界面对应的屏幕元素信息;
    第一关联部分,被配置为基于所述触控操作信息和所述屏幕元素信息,获取所述屏幕元素信息与多个应用的关联度,所述多个应用为同一类型应用;
    第一显示部分,被配置为对所述关联度满足预设条件的应用进行推荐显示。
  30. 一种应用推荐装置,所述装置包括:
    第二获取部分,被配置为获取针对用户界面的触控操作信息,以及获取所述用户界面对应的屏幕元素信息;
    第二关联部分,被配置为基于所述屏幕元素信息和所述触控操作信息进行联合学习,并与预设的样本融合信息和预设的各个应用的样本序列行为信息进行关联处理,获取所述屏幕元素信息与多个应用的关联度;
    第二显示部分,被配置为对所述关联度满足预设条件的应用进行推荐显示。
  31. 一种应用推荐设备,所述应用推荐设备包括存储器和处理器;
    所述存储器存储有可在所述处理器上运行的计算机程序;
    所述处理器执行所述计算机程序时实现权利要求1至25任一项所述的方法,或者权利要求26至28任一项所述的方法。
  32. 一种计算机可读存储介质,其上存储有可执行指令,被配置为被处理器执行时,实现权利要求1至25任一项所述的方法,或者权利要求26至28任一项所述的方法。
PCT/CN2022/120921 2021-11-22 2022-09-23 应用推荐方法、装置、设备和计算机可读存储介质 WO2023087915A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111387371.6A CN113901337A (zh) 2021-11-22 2021-11-22 应用推荐方法、装置、设备和计算机可读存储介质
CN202111387371.6 2021-11-22

Publications (1)

Publication Number Publication Date
WO2023087915A1 true WO2023087915A1 (zh) 2023-05-25

Family

ID=79194840

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/120921 WO2023087915A1 (zh) 2021-11-22 2022-09-23 应用推荐方法、装置、设备和计算机可读存储介质

Country Status (2)

Country Link
CN (1) CN113901337A (zh)
WO (1) WO2023087915A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113901337A (zh) * 2021-11-22 2022-01-07 杭州逗酷软件科技有限公司 应用推荐方法、装置、设备和计算机可读存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105653134A (zh) * 2015-05-02 2016-06-08 Tcl集团股份有限公司 应用切换方法及其系统
CN106339146A (zh) * 2015-07-09 2017-01-18 华为终端(东莞)有限公司 一种推荐应用的方法及装置
CN108470041A (zh) * 2018-02-12 2018-08-31 维沃移动通信有限公司 一种信息搜索方法及移动终端
CN110061908A (zh) * 2019-04-30 2019-07-26 上海连尚网络科技有限公司 应用程序推荐、装置、电子设备及介质
CN113901337A (zh) * 2021-11-22 2022-01-07 杭州逗酷软件科技有限公司 应用推荐方法、装置、设备和计算机可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105653134A (zh) * 2015-05-02 2016-06-08 Tcl集团股份有限公司 应用切换方法及其系统
CN106339146A (zh) * 2015-07-09 2017-01-18 华为终端(东莞)有限公司 一种推荐应用的方法及装置
CN108470041A (zh) * 2018-02-12 2018-08-31 维沃移动通信有限公司 一种信息搜索方法及移动终端
CN110061908A (zh) * 2019-04-30 2019-07-26 上海连尚网络科技有限公司 应用程序推荐、装置、电子设备及介质
CN113901337A (zh) * 2021-11-22 2022-01-07 杭州逗酷软件科技有限公司 应用推荐方法、装置、设备和计算机可读存储介质

Also Published As

Publication number Publication date
CN113901337A (zh) 2022-01-07

Similar Documents

Publication Publication Date Title
CN112685565B (zh) 基于多模态信息融合的文本分类方法、及其相关设备
US11016638B2 (en) Interactive answer boxes for user search queries
US10664504B2 (en) Interaction method and apparatus based on artificial intelligence
CN105320428B (zh) 用于提供图像的方法和设备
CN105706080B (zh) 扩增并呈现捕获的数据
WO2015191652A1 (en) Modeling interestingness with deep neural networks
US11860928B2 (en) Dialog-based image retrieval with contextual information
US10394408B1 (en) Recommending media based on received signals indicating user interest in a plurality of recommended media items
KR20150132860A (ko) 지식으로의 원-스텝 액세스를 제공하는 적응적 사용자 인터페이스를 갖춘 지능형 인터넷 시스템
US10783549B2 (en) Determining persuasiveness of user-authored digital content items
CN110851712B (zh) 书籍信息的推荐方法、设备以及计算机可读介质
CN112948708A (zh) 一种短视频推荐方法
EP4310695A1 (en) Data processing method and apparatus, computer device, and storage medium
US20240114158A1 (en) Hierarchical Video Encoders
WO2023087915A1 (zh) 应用推荐方法、装置、设备和计算机可读存储介质
CN111625715A (zh) 信息提取方法、装置、电子设备及存储介质
CN109101505A (zh) 一种推荐方法、推荐装置和用于推荐的装置
WO2024021685A1 (zh) 回复内容处理方法以及媒体内容互动内容的交互方法
CN116628345A (zh) 一种内容推荐方法、装置、电子设备和存储介质
US20180240023A1 (en) Methods and systems for selecting targeted content by machine learning algorithm
CN114254151A (zh) 搜索词推荐模型的训练方法、搜索词推荐方法和装置
US7909238B2 (en) User-created trade cards
Lee et al. Abnormal usage sequence detection for identification of user needs via recurrent neural network semantic variational autoencoder
CN113806622A (zh) 推荐方法、装置及设备
Martin et al. Fusion-based Representation Learning Model for Multimode User-generated Social Network Content

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22894453

Country of ref document: EP

Kind code of ref document: A1