Server, terminal equipment and resource recommendation method

Abstract

The present disclosure relates to a server, a terminal device and a resource recommendation method, and relates to the field of communication technology. The server includes: a communicator, which is configured to receive input information sent by a terminal device; a controller, which is configured to: determine at least one recommended resource according to the input information received by the communicator, determine a target recommended resource from the at least one recommended resource, input identification information of the target recommended resource into a target language generation model, obtain a first recommended language output by the target language generation model, and send at least one recommended resource and the first recommended language to the terminal device through the communicator. The target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and a standard recommended language corresponding to each identification information.

Description

Technical Field

[0001] The present disclosure relates to the field of communication technology, and in particular to a server, a terminal device, and a resource recommendation method. Background Art

[0002] At present, all smart devices have the function of interacting with users. For example, a user can use voice to send a search request for a certain movie. After the smart TV can recognize the voice content, it will display the resource information of the movie and output a fixed recommendation for the movie. The recommendation is usually based on the fixed setting of the movie and stored in the database. Since the recommendation is stored in advance, the same recommendation may be set for multiple resources due to storage capacity limitations. Therefore, the content of the recommendation output by smart devices is relatively simple and fixed. Summary of the invention

[0003] In order to solve the above technical problems or at least partially solve the above technical problems, the present disclosure provides a server, a terminal device and a resource recommendation method, which are used to solve the problem that the content of the recommendation words output by the current smart devices is relatively single and fixed.

[0004] The technical solutions provided by the embodiments of the present disclosure include:

[0005] In a first aspect, a server is provided, comprising: a communicator configured to receive input information sent by a terminal device;

[0006] The controller is configured as:

[0007] According to input information received by the communicator, at least one recommended resource is determined, a target recommended resource is determined from the at least one recommended resource, identification information of the target recommended resource is input into a target language generation model, a first recommended word output by the target language generation model is obtained, and at least one recommended resource and the first recommended word are sent to a terminal device via the communicator, wherein the target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and a standard recommended word corresponding to each identification information.

[0008] In a second aspect, a terminal device is provided, characterized in that it includes:

[0009] A user interface configured to receive user input;

[0010] An output interface, configured to output user interaction information;

[0011] A communicator, used to communicate with the server;

[0012] The controller is configured to: in response to input received by a user interface, control the communicator to send input information corresponding to the input to a server; the server is configured to, in response to receiving the input information, determine at least one recommended resource according to the input information received by the communicator, determine a target recommended resource from the at least one recommended resource, input the identification information of the target recommended resource into a target language generation model, obtain a first recommended word output by the target language generation model, receive at least one recommended resource and the first recommended word sent by the server through the communicator, and control the output interface to output at least one resource to be recommended and the first recommended word, wherein the target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and a standard recommended word corresponding to each identification information.

[0013] In a third aspect, a resource recommendation method is provided, comprising:

[0014] Receiving input information sent by a terminal device;

[0015] According to the input information, at least one recommended resource is determined; the identification information of the target recommended resource is input into the target language generation model, and the first recommended language output by the target language generation model is obtained; at least one recommended resource and the first recommended language are sent to the terminal device, and the target recommended resource is one of the at least one recommended resource.

[0016] In a fourth aspect, a resource recommendation method is provided, comprising: in response to a user input, sending input information corresponding to the input to a server, wherein the server is configured to, in response to receiving the input information, determine at least one recommended resource corresponding to the input information, input identification information of the target recommended resource into a target language generation model, and obtain a first recommended language output by the target language generation model;

[0017] At least one recommended resource and a first recommended word are received from a server, and an output interface is controlled to output at least one resource to be recommended and the first recommended word, wherein the target recommended resource is one of the at least one recommended resource.

[0018] In a fifth aspect, a resource recommendation system is provided, comprising: a server as in the first aspect and a terminal device as in the second aspect.

[0019] In a sixth aspect, a computer-readable storage medium is provided, comprising: a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the resource recommendation method as shown in the third aspect or the fourth aspect of claim is implemented.

[0020] In the technical solution provided by the embodiment of the present disclosure, when the terminal device receives the user's input, it can send the input information corresponding to the input to the server. The server can determine at least one recommended resource based on the input information corresponding to the input, and generate a recommendation for the target recommended resource in the at least one recommended resource, and send at least one resource and the recommendation of the target recommended resource to the terminal device. In the embodiment of the present disclosure, the recommendation is generated in real time by inputting the label information of the target recommended resource into the target language generation model, and the target language generation model is trained based on the identification information of multiple recommended resources and the standard recommendation corresponding to each identification information. The generated recommendation can be a unique recommendation generated based on the current recommended resource. Compared with the fixed storage recommendation in the prior art, this recommendation generation method can make the generated recommendation more flexible and rich. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] Figure 1 Schematic diagrams of scenes in some embodiments are shown;

[0024] Figure 2 A hardware configuration block diagram of a terminal device in some embodiments is shown;

[0025] Figure 3 A hardware configuration block diagram of a server in some embodiments is shown;

[0026] Figure 4 A diagram showing a software configuration in a terminal device in some embodiments;

[0027] Figure 5 A schematic diagram of a resource recommendation method provided by an embodiment of the present disclosure is shown;

[0028] Figure 6 A schematic diagram of multiple triggering opportunities provided by an embodiment of the present disclosure is shown;

[0029] Figure 7 A schematic diagram of generating a recommendation word by using a target language generation model in an embodiment of the present disclosure is shown;

[0030] Figure 8 A schematic diagram of implementing the recommendation process in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0031] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0032] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.

[0033] Figure 1 Schematic diagram of scenes in some embodiments. Figure 1 As shown, the user can operate the terminal device 200 through the smart device 300 or the control apparatus 100 , and the terminal device 200 performs data communication with the server 400 .

[0034] In some embodiments, the control device 100 may be a remote controller, and the communication between the remote controller and the terminal device includes infrared protocol communication or Bluetooth protocol communication, and other short-range communication methods, and the terminal device 200 is controlled wirelessly or wired. The user can input user commands through buttons on the remote controller, voice input, control panel input, etc. to control the terminal device 200.

[0035] In some embodiments, a smart device 300 (such as a mobile terminal, a tablet computer, a computer, a laptop computer, etc.) may also be used to control the terminal device 200. For example, the terminal device 200 is controlled using an application running on the smart device.

[0036] In some embodiments, the terminal device 200 may not use the above-mentioned smart device or control device to receive instructions, but instead receive user control through touch or gestures.

[0037] In some embodiments, the display device 200 can also be controlled in a manner other than the control device 100 and the smart device 300. For example, the user's voice command control can be directly received through a module for obtaining voice commands configured inside the terminal device 200, or the user's voice command control can be received through a voice control device set outside the terminal device 200.

[0038] In some embodiments, the terminal device 200 may be allowed to communicate via a local area network (LAN), a wireless local area network (WLAN), and other networks. The server 400 may provide various content and interactions to the terminal device 200. The server 400 may be one cluster or multiple clusters, and may include one or more types of servers.

[0039] like Figure 2 As shown, it is a hardware configuration block diagram of a terminal device in some embodiments, the terminal device 200 includes a tuner and demodulator 210, a communicator 220, a detector 230, an external device interface 240, a controller 250, a display 260, an audio output interface 270, a user interface 280, and at least one of a memory and a power supply, wherein the display 260 and the audio output interface 270 are both output interfaces.

[0040] In some embodiments, the controller 250 includes a processor, a video processor, an audio processor, a graphic processor, RAM, ROM, and first to nth interfaces for input / output.

[0041] The display 260 includes a display screen component for presenting images, and a driving component for driving image display, which is used to receive image signals output from the controller, and display video content, image content, and menu control interface components and user control UI interface.

[0042] The display 260 may be a liquid crystal display, an OLED display, or a projection display, and may also be a projection device and a projection screen.

[0043] The communicator 220 is a component for communicating with an external device or server according to various communication protocol types. For example, the communicator may include at least one of a Wifi module, a Bluetooth module, a wired Ethernet module, and other network communication protocol chips or near field communication protocol chips, and an infrared receiver. The terminal device 200 can establish transmission and reception of control signals and data signals with the external control device 100 or the server 400 through the communicator 220.

[0044] The user interface may be used to receive control signals input by the user through the control device 100 (eg, an infrared remote controller, etc.) or through touch or gesture.

[0045] The detector 230 is used to collect signals from the external environment or the external interaction. For example, the detector 230 includes a light receiver, a sensor for collecting the intensity of ambient light; or, the detector 230 includes an image collector, such as a camera, which can be used to collect external environment scenes, user attributes or user interaction gestures; or, the detector 230 includes a sound collector, such as a microphone, etc., for receiving external sounds.

[0046] The external device interface 240 may include, but is not limited to, any one or more of the following: a high-definition multimedia interface (HDMI), an analog or digital high-definition component input interface (component), a composite video input interface (CVBS), a USB input interface (USB), an RGB port, etc. It may also be a composite input / output interface formed by the above multiple interfaces.

[0047] The tuner-demodulator 210 receives broadcast television signals via wired or wireless reception, and demodulates audio and video signals, such as EPG data signals, from a plurality of wireless or wired broadcast television signals.

[0048] In some embodiments, the controller 250 and the tuner-demodulator 210 may be located in different separate devices, that is, the tuner-demodulator 210 may also be located in an external device of the main device where the controller 250 is located, such as an external set-top box.

[0049] The controller 250 controls the operation of the display device and responds to user operations through various software control programs stored in the memory. The controller 250 controls the overall operation of the display device 200. For example, in response to receiving a user command for selecting a UI object to be displayed on the display 260, the controller 250 can perform operations related to the object selected by the user command.

[0050] In some embodiments, the controller includes a central processing unit (CPU), a video processor, an audio processor, a graphics processing unit (GPU), RAM Random Access Memory (RAM), ROM (Read-Only Memory, ROM), a first interface to an nth interface for input / output, a communication bus (Bus), etc.

[0051] The user may input a user command through a graphical user interface (GUI) displayed on the display 260, and the user interface receives the user input command through the graphical user interface (GUI). Alternatively, the user may input a user command through a specific sound or gesture, and the user interface recognizes the sound or gesture through a sensor to receive the user input command.

[0052] "User interface" is the medium interface for interaction and information exchange between applications or operating systems and users. It realizes the conversion between the internal form of information and the form acceptable to users. The commonly used form of user interface is the Graphical User Interface (GUI), which refers to the user interface related to computer operation displayed in a graphical way. It can be an interface element such as an icon, window, control, etc. displayed on the display screen of an electronic device, where the control can include icons, buttons, menus, tabs, text boxes, dialog boxes, status bars, navigation bars, widgets, etc.

[0053] In some embodiments, the terminal device may be a display device, that is, a terminal device with a display function, such as a television, a mobile phone, a computer, a learning machine, etc. In the display device:

[0054] The user interface 280 is configured to receive user input; wherein the input may be voice input, touch input, etc., which is not limited in the embodiments of the present disclosure.

[0055] An output interface (display 260 and / or audio output interface 270), configured to output user interaction information;

[0056] The communicator 220 is used to communicate with the server 400;

[0057] The controller 250 is configured to: in response to the input received by the user interface 280, control the communicator 220 to send input information corresponding to the input to the server 400, and the server 400 is configured to determine at least one recommended resource in response to the input information received by the communicator 220, determine a target recommended resource from the at least one recommended resource, input the identification information of the target recommended resource into the target language generation model, obtain the first recommended language output by the target language generation model, receive at least one recommended resource and the first recommended language sent by the server 400 through the communicator 220, and control the output interface to output at least one resource to be recommended and the first recommended language, wherein the target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and a standard recommended language corresponding to each identification information.

[0058] Controlling the output interface to output at least one resource to be recommended and the first recommendation word refers to outputting at least one resource to be recommended and the first recommendation word through the display 260 and / or the audio output interface 270 .

[0059] In some embodiments, the controller 250 is further configured to:

[0060] The communicator 220 receives ranking information of at least one recommended resource sent by the server 400 ; the ranking information is determined according to the user preference information of the user.

[0061] In some embodiments, the system is further configured to:

[0062] The communicator 220 receives the random resource other than the at least one recommended resource sent by the server 400 .

[0063] In some embodiments, the controller 250 is further configured to:

[0064] Receiving an output mode indication sent by the server 400 through the communicator 220; the output mode indication is used to indicate outputting at least one recommended resource through an interface display mode, and is used to indicate outputting a first recommended phrase through an interface display mode and / or a voice mode;

[0065] The output interface includes: a display 260 and / or an audio output interface 270;

[0066] The controller 250 is specifically configured as follows:

[0067] Controlling the display 260 to output at least one recommended resource and a first recommendation statement in an interface display manner;

[0068] and / or,

[0069] The display 260 is controlled to output at least one recommended resource in an interface display manner, and the audio output interface 270 is controlled to output a first recommended language in a voice manner.

[0070] like Figure 3 FIG. 4 is a block diagram of a hardware configuration of a server in some embodiments; in some embodiments, the server 400 may include:

[0071] The communicator 410 is configured to receive input information sent by the terminal device 200;

[0072] The controller 420 is configured to:

[0073] According to the input information received by the communicator 410, at least one recommended resource is determined, a target recommended resource is determined from the at least one recommended resource, identification information of the target recommended resource is input into a target language generation model, a first recommended word output by the target language generation model is obtained, and at least one recommended resource and the first recommended word are sent to a terminal device via the communicator 410. The target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and a standard recommended word corresponding to each identification information.

[0074] The controller 420 may include a processor, a video processor, an audio processor, a graphics processor, RAM, ROM, etc.

[0075] In some embodiments, the controller 420 is further configured to:

[0076] Obtain recommendation-related factors, which include at least one of the following:

[0077] The current conversation scenario of the user;

[0078] User emotions corresponding to the input information;

[0079] User profile of the user;

[0080] The preference information of the previous context in the current dialogue scene;

[0081] Knowledge graph of target recommendation resources;

[0082] The controller 420 is specifically configured to: input the recommendation-related factors and identification information of the target recommendation resource into the target language generation model, and obtain the first recommendation language of the target recommendation resource output by the target language generation model.

[0083] In some embodiments, the controller 420 is further configured to:

[0084] Obtain identification information of multiple recommended resources and standard recommendation words corresponding to each identification information;

[0085] The identification information of the first recommended resource among multiple recommended resources is input into a preset language generation model, and the recommended words of the first recommended resource output by the preset language generation model are obtained. According to the recommended words of the first recommended resource and the first standard recommended words, the initial language generation model is updated to obtain a target language generation model.

[0086] In some embodiments, the controller 420 is specifically configured to:

[0087] Determine, according to the input information, a correct resource name stored corresponding to the input information, and determine at least one recommended resource based on the correct resource name;

[0088] and / or,

[0089] The input information includes voiceprint information, determining a user profile matching the voiceprint information, and determining at least one recommended resource based on the user profile;

[0090] and / or,

[0091] A target entity is determined based on the input information and contextual information of the user of the input information, and at least one recommended resource associated with the target entity is determined based on the knowledge graph.

[0092] In some embodiments, at least one recommended resource satisfies at least one of the following conditions:

[0093] The search index is greater than or equal to the preset index;

[0094] The resource evaluation parameter is greater than or equal to the preset evaluation parameter;

[0095] The online time is within the preset time range;

[0096] Associated with target regions;

[0097] The user has not played it;

[0098] Not recommended to users.

[0099] In some embodiments, the controller 420 is further configured to:

[0100] estimating a click probability of the at least one recommended resource;

[0101] The at least one recommended resource is sorted according to the click probability to obtain sorting information of the at least one recommended resource; and the sorting information of the at least one recommended resource is sent to the terminal device through the communicator 410.

[0102] In some embodiments, the controller 420 is further configured to: send at least one random resource other than the recommended resource to the terminal device through the communicator 410 .

[0103] In some embodiments, the controller 420 is specifically configured to:

[0104] Obtaining a second recommended language output by the target language generation model, where the second recommended language corresponds to the target recommended resource;

[0105] If it is detected that the fluency parameter of the second recommended word is less than the preset fluency parameter, a third recommended word is determined from the stored recommended words as the first recommended word, and the third recommended word has a similarity parameter with the second recommended word greater than the preset similarity parameter;

[0106] If it is detected that the fluency parameter of the second recommended word is greater than or equal to the preset fluency parameter, the second recommended word is used as the first recommended word.

[0107] In some embodiments, the controller 420 is further configured to:

[0108] An output mode indication is sent to the terminal device via the communicator 410; the output mode indication is used to indicate outputting at least one recommended resource via an interface display mode, and is used to indicate outputting a first recommendation phrase via an interface display mode and / or a voice mode.

[0109] In some embodiments, the terminal device may be a terminal device having an operating system, and the operating system of the terminal device may be an Android operating system, an iOS operating system, etc. The Android operating system is taken as an example for introduction below.

[0110] See also Figure 4 As shown, it is a software configuration diagram of the terminal device in some embodiments. The operating system in the terminal device can be divided into four layers, from top to bottom, namely, the application layer (Applications) layer (referred to as "application layer"), the application framework layer (Application Framework) layer (referred to as "framework layer"), the Android runtime (Android runtime) and system library layer (referred to as "system runtime library layer"), and the kernel layer.

[0111] In some embodiments, at least one application is running in the application layer, and these applications can be window programs, system settings programs, clock programs, etc. provided by the operating system, or applications developed by third-party developers. In specific implementations, the application packages in the application layer are not limited to the above examples.

[0112] The framework layer provides application programming interfaces (APIs) and programming frameworks for applications. The application framework layer includes some predefined functions. The application framework layer is equivalent to a processing center that determines the actions that applications in the application layer take. Through the API interface, applications can access system resources and obtain system services during execution.

[0113] like Figure 4 As shown, in the embodiment of the present application, the application framework layer includes managers, content providers, etc., wherein the manager includes at least one of the following modules: an activity manager (ActivityManager) is used to interact with all activities running in the system; a location manager (Location Manager) is used to provide system services or applications with access to system location services; a package manager (Package Manager) is used to retrieve various information related to the application package currently installed on the device; a notification manager (NotificationManager) is used to control the display and clearing of notification messages; a window manager (Window Manager) is used to manage icons, windows, toolbars, wallpapers, and desktop components on the user interface.

[0114] In some embodiments, the activity manager is used to manage the life cycle of each application and the common navigation back function, such as controlling the exit, opening, and back of the application. The window manager is used to manage all window programs, such as obtaining the display screen size, determining whether there is a status bar, locking the screen, capturing the screen, and controlling the display window changes (for example, reducing the display window, shaking the display, distorting the display, etc.).

[0115] In some embodiments, the system runtime layer provides support for the upper layer, namely the framework layer. When the framework layer is used, the Android operating system will run the C / C++ library contained in the system runtime layer to implement the functions to be implemented by the framework layer.

[0116] In some embodiments, the kernel layer is a layer between hardware and software. Figure 5 As shown, the kernel layer includes at least one of the following drivers: audio driver, display driver, Bluetooth driver, camera driver, WIFI driver, USB driver, HDMI driver, sensor driver (such as fingerprint sensor, temperature sensor, pressure sensor, etc.), and power driver, etc.

[0117] The resource recommendation method provided by the embodiment of the present disclosure can be implemented based on the interaction between the above-mentioned terminal device and the server.

[0118] like Figure 5 FIG. 1 is a schematic diagram of a resource recommendation method provided by an embodiment of the present disclosure; the resource recommendation method includes:

[0119] 501. The terminal device receives user input.

[0120] Among them, the above input is input such as voice input, touch input, etc., which is not limited in the embodiment of the present invention.

[0121] 502. The terminal device sends input information corresponding to the input to the server in response to the user's input.

[0122] In some embodiments, the user's input is voice input, and the terminal device can obtain input information corresponding to the voice input in response to the voice information, and the input information corresponding to the voice input can include voiceprint information corresponding to the voice input, and content information of the voice input, etc. Among them, the terminal device can recognize the voice input to obtain the content information of the voice input and send it to the server.

[0123] In other embodiments, the user's input may be a touch input. In response to the touch input, the terminal device may obtain input content corresponding to the touch input and send the input content to the server.

[0124] In some embodiments, after obtaining input information corresponding to the user's input, the terminal device can perform semantic analysis through a semantic engine to obtain the user's intention.

[0125] In other embodiments, after obtaining the input information of the user's input pair, the terminal device can send it to the server, and the server can perform semantic analysis through semantic inference to obtain the user's intention.

[0126] In some embodiments, throughout the entire voice interaction process, the embodiments of the present disclosure can trigger the resource recommendation process in multiple scenarios, that is, there are multiple triggering opportunities to trigger the execution of subsequent 503 to 507 to enhance the human-computer interaction experience.

[0127] like Figure 6 , which is a schematic diagram of multiple triggering opportunities provided by an embodiment of the present disclosure;

[0128] After obtaining the input information corresponding to the user input, the parsing of the input information can be triggered, such as Figure 6 As shown, there are four kinds of analysis results. Any of the following analysis results can trigger the subsequent resource recommendation process:

[0129] (1) The user has no clear intention;

[0130] In some embodiments, casual queries to the user, or meaningless queries, may be interpreted as the user having no clear intention.

[0131] For example, for Figure 6 As shown in the figure, small talk queries such as "I am bored" and "I am in a bad mood" will be interpreted as the user having no clear intention.

[0132] For example, for Figure 6 Meaningless queries such as "that" shown in the figure will also be interpreted as the user having no clear intention.

[0133] (2) Unable to parse user intent;

[0134] In some embodiments, the user does not express the intention, resulting in the speech engine being unable to parse the user's intention when parsing the input information corresponding to the user's speech input, for example, Figure 6 As shown, for the query "I want", the semantic engine cannot parse the user's intention.

[0135] In some embodiments, the user expresses the intention, but the semantic engine fails to parse it due to positioning, error correction, labeling, and recognition problems of existing semantic engines.

[0136] For example, regarding the positioning problem, some statements from users are new intents and new statements outside the existing fields and intent systems, which cannot be classified into the existing fields and intent systems. Therefore, there will be a problem of being unable to locate the input information of the user, and the user intent cannot be parsed. Exemplarily, in 2020, for the query "epidemic data", at that time, the semantic engine did not cover this newly emerged intent of the epidemic, so the user intent could not be parsed.

[0137] For example, regarding the error correction problem, the input information of the user may not include accurate resource description information, and the user intent cannot be parsed. Exemplarily, for the query "Balala the Fairies", originally the user wanted to search for resources related to "Balala the Fairies", but because the query content is "Balala the Fairies", and the semantic engine cannot recognize "Balala the Fairies" as "Balala the Fairies" through error correction, resulting in the inability to parse the user intent.

[0138] For example, regarding the annotation problem, the input information of the user is the resource name, however, the resource name and resource type are not included in the database corresponding to the semantic engine, and the semantic engine cannot parse the user intent. For example, when the query is "The mighty river flows eastward", the user wants to search for the song "The mighty river flows eastward", but because the resource to which the field "The mighty river flows eastward" belongs and the resource type are not specified in the database corresponding to the semantic engine, the user intent still cannot be parsed.

[0139] For example, regarding the recognition problem, it may be due to incorrect content information recognized when recognizing the voice content, which may also cause the semantic engine to be unable to parse the user intent. For example, for the query "Zhang San's encyclopedia", the user means to view the encyclopedia information of the person Zhang San, but when the recognized content is "Zhang 300's encyclopedia", the semantic engine may not be able to parse the user intent.

[0140] (3) The user has a clear intent, but does not specify a specific resource name;

[0141] For example, for queries such as "I want to watch a movie", "I want to listen to music", "Games", "What are some good movies", etc., the user has stated the intent, but at this time, a specific resource name is not specified.

[0142] In some embodiments, for the situations of (1), (2), and (3) above when triggering the subsequent recommendation process, resource recommendations can be made for the user based on the user portrait formed by the user's historical behavior, the popularity of the resource, the newness of the resource, etc. The specific recommendation strategy can refer to the introduction of any recommendation strategy below and will not be elaborated here.

[0143] Furthermore, the reply content can be output to the user based on the recommended resources. For example, for the above (1) and (2), the voice message "Although I don't understand what you are saying, I guess you like these contents" can be broadcasted. For the case in (3) where the user wants to search for movie resources, the voice message "Although I don't know what movie you want to watch, I guess you will like these movies" can be broadcasted.

[0144] (4) The user specifies the specific resource name.

[0145] In the above situation (4), there may be two situations for the query results of the resource name:

[0146] (4.1) No resources;

[0147] For example, for the query "I want to listen to the fifth piano piece at night", the user clearly stated the resource name "Five piano pieces at night", and the semantic engine can parse the user's intention. However, if the resource is not stored in the music resource library, the query result is no resource.

[0148] In some embodiments, for the above situation (4.1), when the subsequent recommendation process is triggered, resources of the same type as "Night Piano Piece 5" or related resources can be recommended, for example, other piano pieces can be recommended to users. In the recommendation process, resources can also be recommended to users based on knowledge graphs, user portraits formed by user historical behaviors, resource popularity, resource novelty, etc. Specific recommendation strategies can refer to the following introduction to any recommendation strategy, which will not be repeated here.

[0149] (4.2) Have resources.

[0150] For example, for the query "I want to listen to the fifth piano piece at night", if the resource name "the fifth piano piece at night" is spoken, the semantic engine can parse the user's intention and find out if the resource is stored in the music resource library.

[0151] In view of this situation, in the subsequent recommendation process, the resource of "Night Piano Piece 5" that the user has queried can be directly recommended to the user. Further, while presenting the resource of "Night Piano Piece 5" to the user, the same type of resources or related resources can also be recommended to the user.

[0152] In the disclosed embodiment, the resource recommendation process can be triggered for multiple triggering occasions, so that resource recommendation can be carried out throughout the entire user-computer interaction process, thereby improving user experience.

[0153] 503. The server determines at least one recommended resource according to the input information.

[0154] In the embodiments of the present disclosure, the server may determine at least one recommended resource based on one or more recommendation strategies according to the input information.

[0155] In some embodiments, it may include at least one of the following recommendation strategies (A) to (F):

[0156] (A) Recommendation strategy based on query rewriting;

[0157] In some embodiments, according to the input information, the correct resource name saved corresponding to the input information is determined, and at least one recommended resource is determined based on the correct resource name.

[0158] In some embodiments, the mapping pairs of correct queries and incorrect queries may be stored in advance. Exemplarily, as shown in Table 1, it is a schematic diagram of a mapping pair of correct queries and incorrect queries. Assume that the incorrect query input by the user is "Little Pig Peppa" or "Peppa Pig"

[0159]

[0160] Table 1

[0161] For the case where the input information is an incorrect query, the correct query corresponding to the incorrect query (i.e., the above-mentioned correct resource name) can be matched in the mapping pair, and at least one recommended resource is determined based on the correct query.

[0162] In some embodiments, through big data statistical analysis, the mapping pairs of correct queries and incorrect queries obtained based on user log data can be obtained. Due to reasons such as speech recognition errors, incorrect use of the speech function by users, and non-standard user statements, the queries of users are often classified into incorrect business domains, resulting in poor generalization ability and inability to correctly understand the true intentions of users. Query rewriting is to widely explore the statements of users through data mining, and mine similar statements of users under similar intentions from them, so as to rewrite non-standard and incorrect sentences into sentences that are easy to understand and have clear meanings.

[0163] Among them, the process of collecting the set of common related user statements according to user log information includes:

[0164] First, based on big data statistical analysis, analyze the common queries of a large number of users, and use the queries with the most occurrences among users as the set of correct queries. Since these queries are often mentioned by many different users, it can be considered that these are correct queries.

[0165] Then, for the queries of the same user arranged in chronological order, the queries of the user arranged in chronological order can be divided into conversation sessions according to a certain time interval threshold. The time interval threshold can be set according to actual conditions, for example, it can be set to a fixed value of 30 minutes to facilitate recalling related statements within 30 minutes.

[0166] Furthermore, in each session, it is possible to check whether the last query of the session is a correct query. If it is, then the last query is the correct query in this session. The similarity of all other queries in this session to the correct query is calculated. All other queries in this session whose similarity to the correct query is greater than or equal to a preset value can be used as a candidate set of incorrect queries mapped to the correct query. In the above process of determining the set of incorrect queries, some manual screening can be further performed to finally determine the mapping of several incorrect queries to correct queries.

[0167] Among them, you can choose to calculate the similarity of the query by using algorithms such as Euclidean distance, cosine similarity, Jaccard distance, and edit distance. The query mapping data obtained in this way can save the mapping form of the wrong query and the correct query as a recall database for error correction. Recall the correct query corresponding to the original user query.

[0168] The advantage of the recalled queries obtained in the above embodiment is that they are derived from real user queries and new popular queries can be expanded regularly.

[0169] (B) Recommendations based on voiceprint user profiles;

[0170] Among them, the user portrait can be a single-level or multi-level user portrait, and the user portrait can reflect the user's preferences and preferences.

[0171] In some embodiments, the user portrait may include two levels of portrait tags: interest topics and interest tags. For example, interest topics may include: videos, movies, TV series, documentaries, songs, etc. Interest tags may include: science fiction, romance, comedy, war, military, police and gangsters, etc.

[0172] In some embodiments, the input information includes voiceprint information, and a user portrait matching the voiceprint information can be determined, and at least one recommended resource can be determined based on the user portrait. The user portrait can be determined based on the user's historical behavior of playing resources. Specifically, the user portrait can be determined based on the recent user's behavior of playing resources, or the user portrait can be determined based on the historical behavior of users playing resources in the medium and long term. The specific reference period of historical behavior is not limited in the embodiments of the present disclosure, and the user portrait can be determined based on the user's behavior of playing resources within one month, or based on the user's behavior of playing resources within one year.

[0173] In some embodiments, the user profile may be based on user profiles of one or more users of the terminal device.

[0174] In some embodiments, when the voiceprint information is matched with the user portrait, the user portrait of the user who performs voice input to the terminal device can be determined.

[0175] Some terminal devices are shared by multiple users, but the interests of multiple users are different. For example, family members share the same TV, but the interests of family members are different. For such shared terminal devices, if a mixed user profile of multiple people is used to recommend resources to users, the recommended resources may not meet the preferences of a user currently using the terminal device, and the accuracy of resource recommendations is poor. For example: the hostess in the family likes Yi Yang Qianxi, but the host does not like Yi Yang Qianxi, but when the host is watching TV at home, Yi Yang Qianxi's video is recommended to the host, which may lead to the host's poor experience in resource recommendations.

[0176] Since different users have different voiceprint information, in order to achieve accurate recommendations, a user portrait based on personal voiceprint information can be established for each family member based on voiceprint recognition technology. During the recommendation process, the voiceprint information of the current user is identified and matched to the corresponding user portrait, and recommendations are made based on the matched user portrait.

[0177] Exemplarily, as shown in Table 2, there is a schematic diagram of user portraits for different family members; wherein, there are two family members, one member's ID is 001, corresponding to voiceprint information 1, and the other member's ID is 002, corresponding to voiceprint information 2, which correspond to different user portraits, and the user portraits include weights for each type of content of interest.

[0178]

[0179] Table 2

[0180] (C) Knowledge graph association recommendations based on the previous information and entities in the current information (knowledge graph-based recommendations);

[0181] In some embodiments, a target entity can be determined based on input information and user context information of the input information, and at least one recommended resource associated with the target entity can be determined based on the knowledge graph.

[0182] Among them, the knowledge graph is a semantic network that reveals the relationship between entities. On the basis of information, the connection between entities can be established to form a knowledge graph. The knowledge graph can effectively process, handle and integrate complex information into simple and clear "entity, relationship, entity" triples.

[0183] In short, in the above embodiment, the entity involved in the input information and the user's previous information (which may be referred to as the target entity) can be determined, and at least one recommended resource related to the entity can be determined from the pre-formed knowledge graph.

[0184] In some embodiments, an example of knowledge graph association recommendation based on the above information and entities in the current information is provided, assuming that the human-computer interaction dialogue is as follows:

[0185] People: Who is the lead actor in Ruyi’s Royal Love in the Palace?

[0186] Machine: Ruyi's Royal Love in the Palace stars Zhou Xun and Huo Jianhua

[0187] S: I like Zhou Xun the most.

[0188] Machine: Zhou Xun is not only an excellent actress, but also a great singer.

[0189] S: I haven’t heard her songs.

[0190] Machine: I float and you sway, rootless weeds, listen to Zhou Xun's song "Floating" with me!

[0191] From the above human-computer dialogue, we can see that the user mentioned that he / she likes the subject "Zhou Xun" in the above information of the dialogue. At this time, the song resources associated with the subject Zhou Xun can be obtained in the knowledge graph and recommended to the user. For example, in the above human-computer process, the song "Piaoyao" sung by Zhou Xun is recommended to the user.

[0192] (D) Recommendations based on new and popular resources (release time, search index, ratings);

[0193] In some embodiments, resources having a search index greater than or equal to a preset index may be recommended, that is, at least one of the recommended resources needs to satisfy a search index greater than or equal to a preset index.

[0194] In some embodiments, resources whose resource evaluation parameters are greater than or equal to preset evaluation parameters may be recommended, that is, at least one of the recommended resources needs to satisfy the resource evaluation parameter being greater than or equal to the preset evaluation parameter. The resource evaluation parameter may be an evaluation parameter given by a large number of users who have played the resource.

[0195] In some embodiments, resources having a search index greater than or equal to a preset index and a resource evaluation parameter greater than or equal to a preset evaluation parameter may also be recommended.

[0196] (E) Recommendations based on region;

[0197] In some embodiments, resources associated with a target region may be recommended. The target region may be the region where the user is located, or the target region may be a region that the user has paid attention to. For example, if the user is in "Beijing", movies associated with the region information of "Beijing" may be recommended.

[0198] (F) Time-based recommendations.

[0199] In some embodiments, resources whose online time is within a preset time range can be recommended, that is, at least one of the recommended resources needs to meet the requirement of online time being within the preset time range. Wherein, when within the preset time range refers to a period of time closest to the current time, it means that newly launched resources can be recommended. For example, when within the preset time range refers to the month closest to the current time, it means that resources launched within the last month can be recommended.

[0200] Furthermore, in addition to the recommendation strategies mentioned in the above-mentioned embodiments, resources that the user has not played and / or resources that have not been recommended to the user may also be recommended to the user.

[0201] It should be noted that any one or more of the above recommendation strategies can be used alone or in combination, and the embodiments of the present invention do not limit this. By integrating the above multiple function recommendation strategies, resources can be accurately recommended, and the above content in the conversation is combined in the recommendation process, thereby improving the recommendation quality and accuracy.

[0202] 504. The server determines a target recommended resource from at least one recommended resource.

[0203] 505. The server inputs the identification information of the target recommended resource into the target language generation model, and obtains the first recommended language output by the target language generation model.

[0204] The target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources and standard recommended words corresponding to each identification information.

[0205] The identification information of the target recommended resource may be the name of the target recommended resource, relevant production personnel information, type, introduction, comment information, etc. For example, the name of the song is "Dong Feng Po", and the singer is Jay Chou.

[0206] In some embodiments, the identification information of the target recommendation resource may be included in the knowledge graph information of the target recommendation resource.

[0207] Before executing 505 above, the server may also obtain recommendation-related factors, which include but are not limited to at least one of the following:

[0208] 1) The current conversation scenario of the user;

[0209] 2) User emotions corresponding to the input information;

[0210] 3) User profile of the user;

[0211] 4) Preference information of the previous context in the current dialogue scenario;

[0212] 5) Knowledge graph of target recommendation resources;

[0213] In some embodiments, the server may input both the recommendation-related factors and the identification information of the target recommendation resource into the target language generation model, and obtain the first recommendation language of the target recommendation resource output by the target language generation model.

[0214] In some embodiments: identification information of multiple recommended resources and standard recommended words corresponding to each identification information may be obtained from the sample information, and a preset language generation model may be trained once or multiple times according to the identification information of the multiple recommended resources and the standard recommended words corresponding to each identification information to obtain a target language generation model. Specifically, the following steps a to c may be performed once or multiple times to obtain the target language generation model:

[0215] a. inputting identification information of a first recommended resource among multiple recommended resources into a preset language generation model;

[0216] b. Obtaining the recommendation words of the first recommended resource output by the preset language generation model;

[0217] c. Update the initial language generation model according to the recommendation words of the first recommendation resource and the first standard recommendation words.

[0218] In some embodiments, when inputting the identification information of the target recommended resource into the target language generation model, obtaining the first recommended word output by the target language generation model may include: inputting the identification information of the target recommended resource into the target language generation model, obtaining the second recommended word output by the target language generation model, the second recommended word corresponding to the target recommended resource; if it is detected that the fluency parameter of the second recommended word is less than a preset fluency parameter, determining a third recommended word from the stored recommended words as the first recommended word, the third recommended word having a similarity parameter with the second recommended word greater than the preset similarity parameter; if it is detected that the fluency parameter of the second recommended word is greater than or equal to the preset fluency parameter, using the second recommended word as the first recommended word.

[0219] That is to say, when detecting the fluency parameter of the second recommended word, the language model service can be called for the generated second recommended word to obtain the score of the second recommended word to evaluate the fluency of the generated sentence. The language service model is used to evaluate the fluency of the sentence. If the score of the second recommended word is greater than or equal to the preset threshold (i.e., the preset fluency parameter), it can be considered that the fluency of the second recommended word meets the standard and is used as the first recommended word. On the contrary, if the fluency does not meet the standard, the recommended word with the highest similarity to the second recommended word can be determined from the template recommended words in the database as the first recommended word.

[0220] In some embodiments, the doc2vec algorithm can be used to calculate the text vector of the generated sentence, and the cosine similarity between the text vector and the template recommendation in the database can be calculated. If the similarity parameter (i.e., cosine similarity) retrieved is greater than the preset similarity parameter recommendation, the native recommendation is replaced as the final first recommendation; if the similarity parameter (i.e., cosine similarity) is not detected to be greater than the preset similarity parameter recommendation, a template can be randomly selected from the template recommendation in the database according to the resource type as the final first recommendation.

[0221] In order to more clearly illustrate the process of training a target language generation model and generating a recommendation based on the target language generation model in the embodiment of the present disclosure, the following is a process of training a target language generation model and generating a recommendation based on the target language generation model. Figure 7 To explain, Figure 7 This is a schematic diagram of generating recommendation words through a target language generation model in an embodiment of the present disclosure. During the training process of the target language generation model, the target language generation model can be trained in an end-to-end manner by combining dialogue scenarios, user emotions, recommendation basis, knowledge graph information of recommended resources, etc.

[0222] Exemplarily, in the process of obtaining the recommendation words through the target language generation model, the input information may include the following:

[0223] Tag information of the target recommended resource: the name of the target resource. For example Figure 7 The middle one is "blue and white porcelain"

[0224] Conversation scenario: personal or public. The voiceprint information can be used to determine whether the current conversation scenario is a public conversation scenario or a personal conversation scenario. If only one voice input corresponding to a voiceprint information is received within a period of time, it means that it is a personal conversation scenario. If two or more voice inputs corresponding to voiceprint information are received within a period of time, it means that it is a public conversation scenario. Figure 7 The dialogue scenario is not currently taken into account, and this input can be added in actual applications.

[0225] User emotions: The basic forms of user emotions are divided into four types: joy, anger, sadness, and fear. User emotions can be determined by recognizing the tone, speed, and input content of the user's voice input. For example, Figure 7 The user emotion identified in is "sad".

[0226] Recommendation basis: includes the user's user profile, which can be specifically the interest tags in the user profile; preference information in the previous text of the current conversation scenario, for example, the preference entities and preference attribute information mentioned by the user in the previous text of the current conversation scenario. Figure 7 The recommendation is based on "sorrow of separation"

[0227] Knowledge graph information of recommended resources: relevant structured data of the resource. For example, the knowledge graph for music resources may include but is not limited to one or more of the following: music name, singer, lyricist, composer, release time, music style, climax of lyrics, music appreciation). Figure 7 The knowledge graph information of the recommended resources may include: the music name "Blue and White Porcelain", the singer "Jay Chou", the composer "Jay Chou", the lyricist "Fang Wenshan", the release time "2007", the type "Chinese style", the climax of the lyrics "The sky is blue and waiting for the rain and I am waiting for you", and the music appreciation "It's like a green olive in the mouth that can be slowly savored".

[0228] The above input information can be organized into a structured information format required by the model, and input into a trained seq2seq model (i.e., the target language generation model in the embodiment of the present disclosure) to obtain the generated recommendation words.

[0229] Among them, seq2seq belongs to an encoder-decoder structure, in which the encoder encodes the input information and the decoder decodes it to generate the recommendation. Among them, the attention mechanism is also used to improve the model generation effect. Furthermore, in order to ensure that rare words appearing in the names of movies, names of people, song names, etc. can be correctly output, a copy mechanism can also be used to copy rare words directly to the generated recommendation, that is, a de-semantic method is used to copy rare words directly to the generated recommendation.

[0230] Figure 7 In , the target language generation model generates recommendations in an autoregressive manner one word after another. After the two words "listen" and "this song" have been generated, the target language generation model can perform a weighted summation of the probability scores of the words in the attention distribution and the probability scores of the words in the vocabulary distribution to obtain the final probability distribution, and take the word corresponding to the maximum probability in the final profile distribution as the word generated in the current step. Among them, the context vector can also be obtained through the attention distribution, and the weight of the probability score of the word in the attention distribution and the weight of the probability score of the word in the vocabulary distribution can be obtained according to the context vector. Figure 7 The weight of the probability score of a word in the attention distribution is (1-p), and the weight of the probability score of a word in the vocabulary distribution is p. Figure 7 As shown in the figure, the word "Chinese style" corresponds to the maximum probability in the final profile distribution, that is, the word generated in the current step is "Chinese style". After generating words one by one in this way, the final recommendation is "Listening to this Chinese style song Blue and White Porcelain, it is like a green olive in your mouth that you can slowly savor."

[0231] 506. The server sends at least one recommended resource and a first recommendation word to the terminal device.

[0232] In some embodiments: the server may estimate the click probability of at least one recommended resource; sort the at least one recommended resource according to the click probability to obtain sorting information of the at least one recommended resource; and the server may also send the sorting information of the at least one recommended resource to the terminal device.

[0233] The click probability of at least one recommended resource may be estimated by at least one of a deep interest network (DIN), a regression model, and the popularity of the resource. In some embodiments: the server may also send a random resource other than at least one recommended resource to the terminal device.

[0234] 507. The terminal device outputs at least one recommended resource and a first recommendation word.

[0235] The resource recommendation method provided by the embodiment of the present disclosure can send input information corresponding to the input to the server when the terminal device receives the input of the user. The server can determine at least one recommended resource based on the input information corresponding to the input, and generate a recommendation for a target recommended resource in the at least one recommended resource, and send at least one resource and the recommendation for the target recommended resource to the terminal device. In the embodiment of the present disclosure, the recommendation is generated in real time by inputting the identification information of the target recommended resource into a target language generation model, and the target language generation model is trained based on the identification information of multiple recommended resources and the standard recommendation corresponding to each identification information. The generated recommendation can be a unique recommendation generated based on the current recommended resource. Compared with the fixed storage recommendation in the prior art, this recommendation generation method can make the generated recommendation more flexible and rich.

[0236] In some embodiments: an output mode indication is sent to a terminal device through a communicator; the output mode indication is used to indicate outputting at least one recommended resource through an interface display mode, and is used to indicate outputting a first recommended phrase through an interface display mode and / or a voice mode. Accordingly, after the terminal device receives the output mode indication sent by the communicator, the at least one recommended resource and the first recommended phrase can be output according to the output mode indication.

[0237] The output method is crucial to the recommendation system. Traditional recommendation displays are based on text and pictures, which are not vivid enough. For example, the following film and television content is recommended to you, and then the posters of media resources are densely displayed in the interface, which has a poor user experience. The output method in the disclosed embodiment can display the recommendation results through dialogue interaction and interface display, and provide reasons for the recommendation. The reasons for the recommendation can explain the reasons for recommending these resources to users, which can win the trust of users and make it easier for users to accept the recommendation results.

[0238] Among them, one way of explaining the recommendation is to explain the basis of the recommendation to the user based on a preset template, such as explaining to the user the relevant label information of the user portrait on which the recommendation is based.

[0239] Optionally, the recommendation explanation may be included in the first recommendation phrase generated above, or may be independent of the voice content and / or text content of the first recommendation phrase. The recommendation explanation may be displayed in a textual manner on an interface, and / or may be played in a voice manner.

[0240] In order to more clearly illustrate the resource recommendation process in the embodiment of the present disclosure, for example, Figure 8 As shown, the present disclosure provides a schematic diagram of implementing a recommendation process. Figure 8 As can be seen from the figure, the recommendation process is divided into four stages: recall, filtering, sorting, and post-processing.

[0241] In the recall phase, relevant resources can be retrieved from the resource library based on multiple recall strategies and the current scenario. Multiple recall strategies are the recommendation strategies (A) to (E) mentioned above: recommendation based on query rewriting, recommendation based on voiceprint user portrait, recommendation based on knowledge graph, recommendation based on new and hot resources, recommendation based on region, and recommendation based on time. Figure 8 The data warehouse in the computing and storage center stores mapping pairs of incorrect queries and correct queries. These mapping pairs can be mined from user log data in the data source. The mining process can be implemented based on the Spark computing platform.

[0242] In the filtering stage, the large number of resources recalled in the recall stage can be screened to select an appropriate number of resources. For example, resources that the user has viewed and resources that have been recommended to the user can be filtered out.

[0243] In the sorting stage, the sorting strategies include but are not limited to: sorting the possible click probability of users through the deep interest network DIN, regression model, and resource popularity. The deep learning models such as DIN and regression model used in the sorting stage can be trained based on the Tensorflow training platform. The sample data used for training can be stored in Figure 8 The sample data can be stored in the data warehouse in the computing and storage center. Figure 8 The data sources shown are obtained by manually annotating data.

[0244] In the post-processing stage, the sorted resources can be used as resources in the resource library to be recommended to the user (which can be the resource library of the basic data part in the figure), the recommended resources with the top rankings can be presented, and at least one new hot resource can be added. These new hot resources do not have to meet the user's interest preferences. Adding random resources outside the preferences can avoid limiting the user's field of vision. Increase the diversity and novelty of recommendations. Furthermore, a corresponding presentation form can be set for the resources, and a variety of presentation forms can be used for presentation. For example, through the interface display form, combined with voice interaction, the recommendation is presented to the user in the form of a dialogue. The recommendation can be generated by a model for recommendation interpretation (i.e., the target language model in the above embodiment. The model for recommendation interpretation can be trained based on the Tensorflow training platform in the computing and storage center. The sample data that needs to be input during the training process can be obtained from the data warehouse in the computing and storage center. Among them, the sample data in these data warehouses can be data obtained from the data source by manually annotating data.

[0245] against Figure 8The basic data in can provide data support for the above multiple stages. Among them, user behavior data is the historical context behavior data, and user preferences, that is, user portraits, are modeled through these data. Because a terminal device (such as a TV) will be used by multiple members, and the interest portraits of each member are different, it is necessary to use voiceprints to distinguish them, model the preferences of each member, that is, voiceprint user portraits, and store them in voiceprint portrait information. Resources that have been recommended to users can be recorded in historical recommended resources, and resources that have been recommended can no longer be recommended repeatedly.

[0246] In some embodiments, in the process of generating recommendations, not only the identification information of the target recommended resources is considered, but also information such as the dialogue scenario, recommendation basis, user emotions, knowledge graphs, etc. This allows the generated recommendations to meet the scenario requirements, be rich, vivid, and more intelligent.

[0247] In some embodiments, the embodiments of the present disclosure also provide a resource recommendation system, including the server in the embodiments of the present disclosure and the terminal device in the embodiments of the present disclosure.

[0248] In some embodiments, the embodiments of the present disclosure further provide a computer-readable storage medium, including: a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, a resource recommendation method executed by a server is implemented.

[0249] In some embodiments, the embodiments of the present disclosure further provide a computer-readable storage medium, including: a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, a resource recommendation method executed by a terminal device is implemented.

[0250] It should be noted that, in this article, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0251] The above are only specific embodiments of the present disclosure, so that those skilled in the art can understand or implement the present disclosure. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A server, characterized in that: include: A communicator configured to receive input information sent by a terminal device; The controller is configured as: According to the input information received by the communicator, at least one recommended resource is determined, a target recommended resource is determined from the at least one recommended resource, identification information of the target recommended resource is input into a target language generation model, a first recommended language output by the target language generation model is obtained, and the at least one recommended resource and the first recommended language are sent to the terminal device through the communicator; wherein the target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and a standard recommended language corresponding to each identification information; The controller is further configured to: determine the chit-chat queries that appear most frequently as a correct chit-chat query set based on common chit-chat queries of a large number of users; divide the chit-chat queries corresponding to each user according to a certain time interval threshold in chronological order to obtain conversations; for each conversation, when the last chit-chat query in the conversation is a correct chit-chat query, calculate the similarity between other chit-chat queries in the conversation and the last chit-chat query, and when the similarity between other chit-chat queries in the conversation and the last chit-chat query is greater than or equal to a preset value, use the other chit-chat queries as a candidate error set of the last chit-chat query to obtain a mapping pair of the conversation; The controller is specifically configured to: determine the correct resource name stored corresponding to the input information according to the input information and the mapping pair, and determine at least one recommended resource based on the correct resource name.

2. The server according to claim 1, characterized in that: The controller is also configured to: Obtain recommendation-related factors, where the recommendation-related factors include at least one of the following: The current conversation scenario of the user; a user emotion corresponding to the input information; A user profile of the user; The preference information of the previous context in the current dialogue scene; The knowledge graph of the target recommended resource; The controller is specifically configured to: input the recommendation-related factors and the identification information of the target recommendation resource into the target language generation model, and obtain the first recommended language output by the target language generation model.

3. The server according to claim 1 or 2, characterized in that: The controller is specifically configured as follows: The input information includes voiceprint information, determining a user portrait matching the voiceprint information, and determining the at least one recommended resource based on the user portrait; and / or, A target entity is determined according to the input information and user context information of the input information, and at least one recommended resource associated with the target entity is determined based on the knowledge graph.

4. The server according to claim 1 or 2, characterized in that: The at least one recommended resource meets at least one of the following conditions: The search index is greater than or equal to the preset index; The resource evaluation parameter is greater than or equal to the preset evaluation parameter; The online time is within the preset time range; Associated with target regions; The user has not played it; Not recommended to the user.

5. The server according to claim 1 or 2, characterized in that: The controller is also configured to: estimating a click probability of the at least one recommended resource; The at least one recommended resource is sorted according to the click probability to obtain sorting information of the at least one recommended resource; and the sorting information of the at least one recommended resource is sent to the terminal device through the communicator.

6. The server according to claim 1 or 2, characterized in that: The controller is further configured to: send a random resource other than the at least one recommended resource to the terminal device through the communicator.

7. The server according to claim 1 or 2, characterized in that: The controller is specifically configured as follows: Acquire a second recommended word output by the target language generation model, where the second recommended word corresponds to the target recommended resource; If it is detected that the fluency parameter of the second recommended word is less than a preset fluency parameter, determining a third recommended word from the stored recommended words as the first recommended word, wherein the third recommended word has a similarity parameter with the second recommended word greater than a preset similarity parameter; If it is detected that the fluency parameter of the second recommended word is greater than or equal to a preset fluency parameter, the second recommended word is used as the first recommended word.

8. The server according to claim 1 or 2, characterized in that: The controller is also configured to: An output mode indication is sent to the terminal device through the communicator; the output mode indication is used to indicate outputting the at least one recommended resource through an interface display mode, and is used to indicate outputting the first recommended language through an interface display mode and / or a voice mode.

9. A terminal device, characterized in that: include: A user interface configured to receive user input; An output interface, configured to output user interaction information; A communicator, used to communicate with the server; The controller is configured to: in response to the input received by the user interface, control the communicator to send input information corresponding to the input to the server, the server is configured to determine at least one recommended resource according to the input information received by the communicator, determine a target recommended resource from the at least one recommended resource, input the identification information of the target recommended resource into the target language generation model, obtain the first recommended language output by the target language generation model, receive the at least one recommended resource and the first recommended language sent by the server through the communicator, and control the output interface to output the at least one resource to be recommended and the first recommended language, wherein the target language generation model is a model obtained by training a preset language generation model according to sample information, and the sample information includes: identification information of multiple recommended resources, and corresponding to each identification information The server is further configured to: determine the most frequently appearing chat queries as a set of correct chat queries based on common chat queries of a large number of users; divide the chat queries corresponding to each user according to a certain time interval threshold in chronological order to obtain individual conversations; for each conversation, when the last chat query in the conversation is a correct chat query, calculate the similarity between other chat queries in the conversation and the last chat query, and when the similarity between other chat queries in the conversation and the last chat query is greater than or equal to a preset value, use the other chat queries as a candidate error set of the last chat query to obtain a mapping pair of the conversation; the server is specifically configured to: determine the correct resource name saved corresponding to the input information according to the input information and the mapping pair, and determine at least one recommended resource based on the correct resource name.

10. The terminal device according to claim 9, characterized in that: The controller is further configured to: The communicator receives the ranking information of the at least one recommended resource sent by the server; the ranking information is determined according to the user preference information of the user.

11. The terminal device according to claim 9, characterized in that: Also configured as: The communicator receives, through the communicator, at least one random resource other than the recommended resource sent by the server.

12. The terminal device according to claim 9, characterized in that: The controller is also configured to: receiving, through the communicator, an output mode instruction sent by the server; the output mode instruction is used to instruct outputting the at least one recommended resource through an interface display mode, and is used to instruct outputting the first recommended language through an interface display mode and / or a voice mode; The output interface includes: a display and / or an audio output interface; The controller is specifically configured as follows: Controlling the display to output the at least one recommended resource and the first recommendation language in an interface display manner; and / or, The display is controlled to output the at least one recommended resource in an interface display manner, and the audio output interface is controlled to output the first recommended language in a voice manner.

13. A resource recommendation method, characterized in that: include: Receiving input information sent by a terminal device; Determining at least one recommended resource based on the input information; Determine a target recommended resource from the at least one recommended resource, input identification information of the target recommended resource into a target language generation model, and obtain a first recommended language output by the target language generation model; Sending the at least one recommended resource and the first recommended language to the terminal device; Before determining at least one recommended resource according to the input information, the method further includes: determining the chat queries with the highest number of occurrences as a correct chat query set based on common chat queries of a large number of users; dividing the chat queries corresponding to each user according to a certain time interval threshold in chronological order to obtain conversations; for each conversation, when the last chat query of the conversation is a correct chat query, calculating the similarity between other chat queries in the conversation and the last chat query, and when the similarity between other chat queries in the conversation and the last chat query is greater than or equal to a preset value, using the other chat queries as a candidate error set of the last chat query to obtain a mapping pair of conversations; Determining at least one recommended resource according to the input information includes: determining a correct resource name stored corresponding to the input information according to the input information and the mapping pair, and determining at least one recommended resource based on the correct resource name.

14. A resource recommendation method, characterized in that: include: In response to the user's input, the input information corresponding to the input is sent to the server, wherein the server is configured to determine at least one corresponding recommended resource according to the input information, determine a target recommended resource from the at least one recommended resource, input the identification information of the target recommended resource into the target language generation model, and obtain the first recommended language output by the target language generation model, wherein the target language generation model is a model obtained by training a preset language generation model according to sample information, wherein the sample information includes: identification information of multiple recommended resources, and a standard recommended language corresponding to each identification information; the server is further configured to: based on common chat inquiries of a large number of users, determine the chat inquiries that appear the most times as correct A chat query set; in chronological order, the chat queries corresponding to each user are divided according to a certain time interval threshold to obtain individual conversations; for each conversation, when the last chat query in the conversation is a correct chat query, the similarity between other chat queries in the conversation and the last chat query is calculated, and when the similarity between other chat queries in the conversation and the last chat query is greater than or equal to a preset value, the other chat queries are used as a candidate error set of the last chat query to obtain a mapping pair of the conversation; the server is specifically configured to: determine the correct resource name stored corresponding to the input information according to the input information and the mapping pair, and determine at least one recommended resource based on the correct resource name; The at least one recommended resource and the first recommended word sent by the server are received, and an output interface is controlled to output the at least one resource to be recommended and the first recommended word.

15. A resource recommendation system, characterized in that: include: A server as claimed in any one of claims 1 to 8 and a terminal device as claimed in any one of claims 9 to 12.

16. A computer-readable storage medium, characterized in that: include: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the resource recommendation method according to claim 13 or 14 is implemented.

Images