A scene-adaptive-based emoticon recommendation method and system

Abstract

The application discloses a kind of expression symbol recommendation method and system based on scene adaptation, the text keyword corresponding to its determination by input voice information, then according to the current scene identification of several current scene identifications detected, obtain current scene identification result in combination with preset neural network model, then by the first scene record corresponding to the scene identification result is obtained by matching the scene identification result to the preset several scene records, according to the text keyword, current scene identification result and the first scene record, in combination with preset scene adaptation algorithm, obtain recommended expression symbol list, support according to voice recommendation expression symbol, improve the accuracy of expression symbol recommendation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, in particular to an expression symbol recommendation method and system based on scene adaptation. BACKGROUND

[0002] With the development of Internet technology, using instant messaging tools has become an indispensable way of communication for people, and the forms of chat are also diverse, mainly including text, expression, voice, video, etc. Among them, expression is the most expressive form of user's mood, which can express many meanings that cannot be expressed by text.

[0003] At present, when a user needs to send an expression during chatting, the user needs to open the expression selection page and view and select the expression needed by the user in sequence according to the existing expression order. If the expression selection page of the current terminal device contains a large number of expressions, and the expression needed by the user is relatively late. The user needs to constantly page and find, which increases the operation complexity of the user and the time needed to select the expression.

[0004] At the same time, with the maturity of speech recognition technology and the increase of mobile terminal user groups, the input method of converting voice input into text on mobile terminals has been popularized. However, the current input method has the following problems: only supports voice input into text and punctuation marks, does not support recognizing scenes, contexts and automatically recommending expression symbols in voice information; the existing expression symbol recommendation sorting does not support scene adaptation, resulting in low recommendation accuracy. SUMMARY

[0005] The present application provides an expression symbol recommendation method and system based on scene adaptation, which supports recommending expression symbols according to voice and improves the accuracy of expression symbol recommendation.

[0006] In order to achieve the above technical purpose, the present application provides an expression symbol recommendation method based on scene adaptation, comprising:

[0007] determining the text keyword corresponding to the input voice information;

[0008] obtaining the current scene recognition result in combination with the preset neural network model according to the detected several current scene identifiers;

[0009] matching the scene recognition result by the preset several scene records to obtain the first scene record corresponding to the scene recognition result;

[0010] obtaining the recommended expression symbol list in combination with the preset scene adaptation algorithm according to the text keyword, the current scene recognition result and the first scene record.

[0011] The application provides a scene-adaptive emoticon recommendation method, which comprises the following steps: firstly, determining the text corresponding to the input voice, determining the keywords thereof, and facilitating subsequent searching; then, obtaining the current scene recognition result according to the detected several current scene identifiers through a preset neural network model, improving the accuracy of scene detection; according to the detected scene recognition result, finding the corresponding first scene record by matching the pre-stored scene record, improving the accuracy of recommendation; then, according to the obtained text keywords, the current scene recognition result and the first scene record, the emoticon conforming to the current scene and the input voice is obtained through a preset scene-adaptive algorithm, and the accuracy of recommendation is improved.

[0012] As a preferred example, in the training process of the preset neural network model, the following steps are included:

[0013] Receiving the new neural network model sent by the server end in time, saving the new neural network model to the user end locally, and deleting the old neural network model;

[0014] The server end updates the neural network model according to the new neural network model, and the user end receives and saves the new neural network model through network connection, so as to replace the old neural network model;

[0015] The scene record comprises scene identifiers, historical use emoticons, text keywords and time periods.

[0016] The application trains the preset neural network model on the server end, avoids the waste of space of the user end in the training process, and improves the accuracy of scene recognition of the neural network model by updating the neural network model according to the saved scene record in time on the server end, and then sending the neural network model to the user end through network connection, so that the user end does not need to train and download, and the detection speed is improved.

[0017] As a preferred example, in the step of obtaining the current scene recognition result according to the detected several current scene identifiers, the following steps are included:

[0018] Obtaining several first scene identifiers in the current area through detection, wherein the several first scene identifiers comprise SSIDs, GPS positioning and mobile network positioning;

[0019] Screening the obtained several first scene identifiers through a preset scene identifier screening condition to obtain several current scene identifiers;

[0020] The scene identifier screening condition comprises identifier words, set geographical area range and signal strength.

[0021] The application improves the comprehensiveness of scene recognition by detecting a plurality of first scene identifiers including SSID, GPS positioning and mobile network positioning and the like, and then screening the obtained scene identifiers to further obtain effective scene identifiers, thereby improving the accuracy of scene recognition.

[0022] As a preferred example, the matching of the scene recognition result with the corresponding first scene record is performed by presetting a plurality of scene records, and specifically includes:

[0023] The method for matching the scene record by the user terminal is determined by judging whether the current user terminal is in a network connection state.

[0024] If the user terminal is in a network connection state, the scene recognition result is sent to the server terminal, and the first scene record matched with the scene recognition result is obtained from the server terminal according to a plurality of scene records pre-stored in the server terminal and is sent to the user terminal.

[0025] If the user terminal is not in a network connection state, the first scene record matched with the scene recognition result is obtained from a plurality of scene records saved in the user terminal.

[0026] In the process of matching the scene record, it is firstly judged whether the current user terminal is in a network connection state, and then it is determined which matching method is used, thereby ensuring the use of the current user. If the current user terminal is in a network connection state, the current scene recognition result can be sent to the server terminal for matching the scene record, which reduces the resource waste of the user terminal and improves the accuracy of detection. If the current user terminal is not in a network connection state, the scene record saved in the current user terminal can also be matched, thereby ensuring that the user can use the recommended method of the application.

[0027] As a preferred example, the recommended emoticon is recommended by a preset scene adaptive algorithm according to the text keyword, the current scene recognition result and the first scene record, and specifically includes:

[0028] The recommended emoticon is recommended by a preset scene adaptive algorithm according to the text keyword, the current scene recognition result and the first scene record, and specifically includes:

[0029] The recommended emoticon list is dynamically updated in the recommended order of the emoticon list by a preset time threshold and the text keyword corresponding to the input voice obtained by continuously updating the user terminal.

[0030] The preset sorting mode includes a frequency of use of an emoji in the scene record and a set threshold of emoji recommendation.

[0031] The application obtains a recommended emoji list according to set emoji recommendation information by a preset scene adaptive algorithm, combines previously obtained text keywords, a current scene recognition result and corresponding scene records, and improves the accuracy of recommendation.

[0032] As a preferred example, the obtaining of the recommended emoji list specifically includes:

[0033] When the recommended emoji list is obtained, the user terminal saves a user-selected emoji, a text keyword corresponding to the voice, a current scene recognition result and other information as scene records in the local terminal;

[0034] The user terminal sends the saved scene records to the server terminal through a network connection;

[0035] The server terminal updates scene records corresponding to the information in the database according to the information in the scene records sent by the user terminal.

[0036] The application saves a user-selected emoji, a text keyword corresponding to the voice, a current scene recognition result and other information as scene records in the local terminal after each recommendation, and sends the information to the server terminal, so that the scene records of the server terminal can be updated according to the information, thereby further improving the accuracy of recommendation.

[0037] As a preferred example, before the text keyword corresponding to the input voice information is determined, the method specifically includes:

[0038] According to a user terminal emoji information setting interface, emoji information is set; the emoji information includes an emoji recommendation threshold and an emoji selection mode; the emoji recommendation threshold includes an emoji recommendation number and a recommendation priority level; and the emoji selection mode includes clicking, shaking and staring;

[0039] The user terminal saves the set emoji information to the local terminal.

[0040] The application provides an emoticon information setting interface, so that a user can set recommended information of emoticons according to his own needs, the recommended information including an emoticon recommendation threshold and an emoticon selection mode, the emoticon recommendation threshold including an emoticon recommendation number and a recommendation priority level, and the emoticon selection mode including clicking, shaking and staring, so that multiple selection modes are provided when selecting recommended emoticons, the flexibility is high, and the user experience is improved.

[0041] In another aspect, the application provides an emoticon recommendation system based on scene adaptation, including a text analysis module, a scene recognition module, a scene detection module and a recommendation module.

[0042] The text analysis module is used for determining the corresponding text keyword of the input voice information.

[0043] The scene recognition module is used for obtaining the current scene recognition result in combination with a preset neural network model according to the detected current scene identifiers.

[0044] The scene detection module is used for matching the scene recognition result with a preset scene record to obtain a first scene record corresponding to the scene recognition result.

[0045] The recommendation module is used for obtaining a recommended emoticon list in combination with a preset scene adaptation algorithm according to the text keyword, the current scene recognition result and the first scene record.

[0046] The application provides an emoticon recommendation system based on scene adaptation, which first determines the text corresponding to the input voice and the keyword thereof through a text analysis module, facilitates subsequent searching, then obtains the current scene recognition result in combination with a preset neural network model according to the detected current scene identifiers through a scene recognition module, improves the accuracy of scene detection, then finds the corresponding first scene record by matching the preset scene record according to the detected scene recognition result through a scene detection module, improves the accuracy of recommendation, and finally obtains the emoticon conforming to the current scene and the keyword of the input voice through a preset scene adaptation algorithm according to the obtained text keyword, the current scene recognition result and the first scene record through a recommendation module, and improves the accuracy of recommendation.

[0047] As a preferred example, the scene recognition module includes a training unit and a screening unit.

[0048] The training unit is used for receiving a new neural network model sent by a server in a timing manner, saving the new neural network model to a user terminal, and deleting an old neural network model; the server trains the neural network model in a timing manner according to a plurality of scene records stored in the server, obtains a new neural network model, and sends the new neural network model to the user terminal; the scene record includes a scene identifier, a history use emoticon, a text keyword, and a time period.

[0049] The screening unit is used for detecting a plurality of first scene identifiers in a current area, screening the plurality of first scene identifiers according to a preset scene identifier screening condition, and obtaining a plurality of current scene identifiers; the scene identifier screening condition includes an identifier text, a set geographical area range, and a signal strength.

[0050] The application trains a preset neural network model on a server through a training unit, avoids wasting space of a user terminal in a training process, and improves the accuracy of scene recognition of the neural network model by updating the neural network model on the server in a timing manner according to a stored scene record, and then sending the neural network model to the user terminal through a network connection, without training and downloading of the user terminal, thereby improving detection speed, obtaining a plurality of first scene identifiers through a screening unit, and improving the comprehensiveness of scene recognition, and then screening the obtained scene identifiers to further obtain effective scene identifiers, thereby improving the accuracy of scene recognition.

[0051] As a preferred example, the scene detection module includes a judgment unit and a detection unit.

[0052] The judgment unit is used for judging whether a current user terminal is in a network connection state, and determining a method for matching the scene record by the user terminal.

[0053] The detection unit is used for sending the scene recognition result to the server when the user terminal is in the network connection state, receiving a first scene record matched with the scene recognition result obtained by the server according to a plurality of scene records prestored in the server, and obtaining the first scene record matched with the scene recognition result according to a plurality of scene records stored in the user terminal when the user terminal is not in the network connection state.

[0054] In the process of carrying out the matching of the scene record, first, the judgment unit judges whether the current user terminal is in the network connection state, and then decides to use which matching method, ensures the use of the current user, and then the detection unit judges that if the current user terminal is in the network connection state, the current scene recognition result can be sent to the server terminal for the matching of the scene record, on the one hand, reduces the resource waste of the user terminal, and on the other hand, improves the accuracy of detection. If the current user terminal is not in the network connection state, the scene record saved in the current user terminal can also be matched, so that the user can use the recommended method of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0055] Figure 1 : a flowchart of an expression symbol recommendation method based on scene adaptation provided by an embodiment of the present application;

[0056] Figure 2 : a structural diagram of an expression symbol recommendation system based on scene adaptation provided by an embodiment of the present application;

[0057] Figure 3 : a network architecture diagram of an expression symbol recommendation system based on scene adaptation provided by another embodiment of the present application;

[0058] Figure 4 : another flowchart of an expression symbol recommendation method based on scene adaptation provided by another embodiment of the present application;

[0059] Figure 5 : a flowchart of an online expression symbol recommendation method provided by another embodiment of the present application;

[0060] Figure 6 : a flowchart of an offline expression symbol recommendation method provided by another embodiment of the present application. DETAILED DESCRIPTION

[0061] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0062] Embodiment one

[0063] Please refer to Figure 1 A flowchart of an expression symbol recommendation method based on scene adaptation provided by an embodiment of the present application mainly includes steps 101 to 104, which specifically include:

[0064] Step 101: determining the corresponding text keyword according to the input voice information.

[0065] In this embodiment, this step specifically includes: converting the input voice into text, and obtaining the keyword of the text.

[0066] Step 102: obtaining the current scene recognition result according to the detected several current scene identifiers, and combining the preset neural network model.

[0067] In this embodiment, before this step is implemented, the training process of the neural network model includes: receiving the new neural network model sent by the server side in a timely manner, saving the new neural network model to the user side locally, and deleting the old neural network model; the server side trains the neural network model according to the several scene records saved on the server side in a timely manner, obtains a new neural network model, and sends the new neural network model to the user side; the scene record includes scene identifier, historical use of emoticon, text keyword and time period.

[0068] In this step of this embodiment, the detected several current scene identifiers include: obtaining several first scene identifiers in the current area through detection, and the several first scene identifiers include SSID, GPS positioning and mobile network positioning; the several first scene identifiers obtained are screened through the preset scene identifier screening condition to obtain several current scene identifiers; the scene identifier screening condition includes identifier text, set geographical area range and signal strength.

[0069] Step 103: matching the scene recognition result through the preset several scene records to obtain the first scene record corresponding to the scene recognition result.

[0070] In this embodiment, this step specifically includes: determining the method of matching the scene record of the user side by judging whether the current user side is in a network connection state; if the user side is in a network connection state, sending the scene recognition result to the server side to receive the first scene record matched with the scene recognition result obtained by the server side according to the several scene records pre-stored in the server side; if the user side is not in a network connection state, obtaining the first scene record matched with the scene recognition result according to the several scene records saved in the user side.

[0071] Step 104: obtaining the recommended emoticon list according to the text keyword, the current scene recognition result and the first scene record, and combining the preset scene adaptation algorithm.

[0072] In the embodiment, the step specifically includes: using a preset scene adaptation algorithm to obtain a recommended emoticon list by selecting from a local emoticon library according to the text keyword, the current scene recognition result and the first scene record, and using a preset sorting mode to display the recommended emoticon list; the recommended emoticon list is dynamically updated in the recommended order of the emoticon list by using a preset time threshold and the text keyword corresponding to the input voice obtained by the user terminal continuously; and the preset sorting mode includes a use frequency of the emoticon in the scene record and a set emoticon recommendation threshold.

[0073] Please refer to Figure 2 A structure diagram of an emoticon recommendation system based on scene adaptation provided by the embodiment of the present application mainly includes a text analysis module 201, a scene recognition module 202, a scene detection module 203 and a recommendation module 204.

[0074] The text analysis module 201 is used to determine the text keyword corresponding to the input voice information.

[0075] The scene recognition module 202 is used to obtain a current scene recognition result by combining a preset neural network model according to the detected current scene identifier.

[0076] The scene detection module 203 is used to obtain a first scene record corresponding to the scene recognition result by matching the scene recognition result with a preset scene record.

[0077] The recommendation module 204 is used to obtain a recommended emoticon list by combining a preset scene adaptation algorithm according to the text keyword, the current scene recognition result and the first scene record.

[0078] In the embodiment, the scene recognition module 202 includes a training unit and a screening unit.

[0079] The training unit is used to receive a new neural network model sent by a server terminal at a regular time, save the new neural network model to a local user terminal and delete an old neural network model; the server terminal trains the neural network model according to a plurality of scene records saved in the server terminal at a regular time, obtains a new neural network model and sends the new neural network model to the user terminal; and the scene record includes a scene identifier, a historical use emoticon, a text keyword and a time period.

[0080] The screening unit is used for detecting a plurality of first scene identifications in a current area, the plurality of first scene identifications including a SSID, a GPS positioning and a mobile network positioning; screening the plurality of first scene identifications obtained through a preset scene identification screening condition, obtaining a plurality of current scene identifications; the scene identification screening condition including an identification word, a set geographical area range and a signal strength.

[0081] In the embodiment, the scene detection module 203 includes a judging unit and a detection unit.

[0082] The judging unit is used for judging whether a current user terminal is in a network connection state, and determining a method of matching the user terminal with the scene record.

[0083] The detection unit is used for, when the user terminal is in the network connection state, sending the scene recognition result to a server terminal, receiving a first scene record matched with the scene recognition result obtained by the server terminal according to a plurality of scene records pre-stored in the server terminal; and when the user terminal is not in the network connection state, obtaining a first scene record matched with the scene recognition result according to a plurality of scene records saved in the user terminal.

[0084] Embodiment two

[0085] Please refer to Figure 3 A network architecture diagram of an expression symbol recommendation system based on scene adaptation provided by another embodiment of the present application mainly includes:

[0086] The network architecture includes a server SVR located in the Internet, a plurality of mobile hotspots and a mobile terminal CLT, the mobile terminal CLT communicates with the server SVR, a current area where the mobile terminal CLT is located is a family residence, and hot spot identification information SSID of the area includes: {Tom's home}, {Tom's neighbor's home} and {a convenience store}.

[0087] Please refer to Figure 4 Another flowchart of an expression symbol recommendation method based on scene adaptation provided by another embodiment of the present application. It mainly includes steps 401 to 408, mainly including:

[0088] Step 401: the user sets expression symbol information and saves it to the local.

[0089] In the embodiment, the step specifically includes: entering an emoji information setting interface of the user terminal CLT, setting a recommended geographic area range as a company range of the user Tom, screening a voice input recommended emoji function, searching a geographic area radius of 300 meters, setting a recommended emoji quantity of a matching scene as 6, setting a recommended priority level as a priority recommendation of a historically frequently used emoji in the scene, setting an emoji selection mode as a shaking selection of an emoji, and specifically including a shaking of one time to add a currently selected emoji and a shaking of two times to switch a next selected emoji. The user terminal CLT saves the emoji information set by the user Tom to the local.

[0090] Step 402: The server trains a neural network model and saves the model to the local.

[0091] In the embodiment, the step specifically includes: the server SVR trains a neural network model by using existing scene records in the database. When the user terminal CLT has a network connection, the server SVR sends the trained neural network model to the user terminal CLT and saves the model to the local.

[0092] Step 403: The user opens a voice input function.

[0093] In the embodiment, the step specifically includes: the user Tom opens the voice input function by using the user terminal CLT.

[0094] Step 404: The user terminal checks whether a network is connected.

[0095] In the embodiment, the step specifically includes: the user terminal CLT checks whether a network is connected currently. If yes, the step jumps to step 405. If no, the step jumps to step 406.

[0096] Step 405: An online emoji recommendation method is entered.

[0097] Step 406: An offline emoji recommendation method is entered.

[0098] Step 407: The user terminal saves a scene record to the local and sends the record to the server to save to the database.

[0099] In the embodiment, the step specifically includes: after the voice input is ended, the user terminal CLT saves the selected emoji, the corresponding keyword, the geographic information, the time period and other information of the user Tom as a scene record to the local, sends the scene record to the server SVR when a network is connected, and saves the record to the database.

[0100] Step 408: The server updates the scene record and re-trains a neural network model and saves the model to the local.

[0101] In the embodiment, the step specifically includes: the server SVR regularly counts the frequency of the user TOM using the emoticon in a specific scene, the time period of use and other information, and updates the corresponding scene record in the database accordingly, and regularly re-trains the neural network model using the existing scene record in the database. When the user terminal CLT has a network connection, the server SVR sends the trained neural network model to the user terminal CLT and saves it locally.

[0102] Please refer to Figure 5 An online emoticon recommendation method flowchart provided by the embodiment of the application mainly includes steps 501 to 509, and specifically includes:

[0103] Step 501: Enter the online emoticon recommendation method.

[0104] In the embodiment, the step specifically includes: the user terminal CLT detects that there is a network connection at present, and enters the online emoticon recommendation method process.

[0105] Step 502: The user terminal detects the scene identifier.

[0106] In the embodiment, the step specifically includes: the user terminal CLT detects that there are a plurality of scene identifiers such as SSIDs, GPS positioning, mobile network positioning in the current area. The user Tom is currently connected to a mobile hotspot with the SSID {Tom home}, and the signal strength of the mobile hotspot is strong; the SSIDs detected in the search geographic area range set by the user Tom also include: {Tom neighbor's home}, {a convenience store}, wherein the signal strength of the mobile hotspot with the SSID {Tom neighbor's home} is general, and the signal strength of the mobile hotspot with the SSID {a convenience store} is weak.

[0107] Step 503: The user terminal inputs the effective identifier into the neural network model to obtain the scene recognition result.

[0108] In the embodiment, the step specifically includes: the user terminal CLT inputs the effective identifier screened into the local neural network model, and obtains the scene recognition result output by the neural network model as: the user terminal CLT has a 90% possibility of being located in a residential area and a 10% possibility of being located in a store area, and the neural network model judges that the user terminal CLT is currently located in a residential area.

[0109] Step 504: The user terminal sends the current scene to the server.

[0110] In the embodiment, the step specifically includes: the user terminal CLT logs in the server SVR, and sends the current scene to the server SVR.

[0111] Step 505: The server searches for the corresponding scene record.

[0112] In the embodiment, the step specifically includes that the server SVR searches the database for the corresponding scene record, and if there is no corresponding scene record, the process jumps to step 507.

[0113] Step 506: The server sends the scene record to the user terminal and saves it locally.

[0114] In the embodiment, the step specifically includes that the server SVR sends the scene record containing the valid scene identifier, the historical use of the emoji, the keyword corresponding to the emoji, the geographic information, the time period, the activity track and the like to the user terminal CLT, and the user terminal CLT saves the scene record locally.

[0115] Step 507: The user terminal recommends the emoji.

[0116] In the embodiment, the step specifically includes that the user terminal CLT selects the corresponding emoji from the local emoji library according to the voice information keyword input by the user Tom, the current scene, the scene record (if any), the emoji recommendation threshold and the current time, and displays the emoji recommendation list to the user Tom after sorting according to the historical use of the emoji frequency in the scene record and the emoji recommendation threshold set by the user Tom.

[0117] Step 508: The user selects the emoji as the associated input.

[0118] In the embodiment, the step specifically includes that the user Tom shakes the mobile terminal to select the appropriate emoji as the associated input of the voice keyword.

[0119] Step 509: The user terminal updates the emoji recommendation order.

[0120] In the embodiment, the step specifically includes that the user terminal CLT periodically updates the emoji recommendation order.

[0121] Please refer to Figure 6 , a flowchart of an offline emoji recommendation method provided by the embodiment of the present application, which mainly includes steps 601 to 607, and specifically includes:

[0122] Step 601: Enter the offline emoji recommendation method.

[0123] In the embodiment, the step specifically includes that the user terminal CLT detects that there is no network connection, and enters the offline emoji recommendation method process.

[0124] Step 602: The user terminal detects the scene identifier.

[0125] In the embodiment, the step specifically includes: detecting, by the user terminal CLT, a plurality of scene identifiers such as SSIDs, GPS positioning, mobile network positioning, and the like in the current area.

[0126] Step 603: inputting, by the user terminal, the valid identifiers into the neural network model to obtain a scene recognition result.

[0127] In the embodiment, the step specifically includes: inputting, by the user terminal CLT, the valid identifiers screened to the local neural network model to obtain a scene recognition result output by the neural network model, that is, the user terminal CLT has a 60% possibility of being located in a home area and a 40% possibility of being located in a store area, and the neural network model determines that the current scene of the user terminal CLT is located in the home area.

[0128] Step 604: detecting, by the user terminal, a local scene record.

[0129] In the embodiment, the step specifically includes: detecting, by the user terminal CLT, whether there is a scene record in the local area.

[0130] Step 605: recommending, by the user terminal, an emoji.

[0131] In the embodiment, the step specifically includes: selecting, by the user terminal CLT, corresponding emojis from a local emoji library according to the voice information keyword input by the user Tom, the current scene, the scene record (if any), the emoji recommendation threshold, the current time, and the like, and displaying an emoji recommendation list to the user Tom in a sequence according to the historical use frequency of the emojis in the scene record, the emoji recommendation threshold set by the user Tom, and the like.

[0132] Step 606: selecting, by the user, an emoji as an associated input.

[0133] In the embodiment, the step specifically includes: shaking the mobile terminal by the user Tom to select a suitable emoji as an associated input of the voice keyword.

[0134] Step 607: updating, by the user terminal, an emoji recommendation sequence.

[0135] In the embodiment, the step specifically includes: periodically updating, by the user terminal CLT, the emoji recommendation sequence.

[0136] In the embodiment two provided by the application, based on the network architecture diagram provided by the embodiment of the application, a whole route diagram of the expression symbol recommendation method based on scene adaptation is provided. In the embodiment, the user sets the expression symbol information and saves it to the local, sets according to the own demand, improves the user experience, then the server trains the neural network model and saves it to the local, saves the time of training the network model on the user side, improves the recommendation speed, then judges whether the current user side is in the network connection state, provides a plurality of expression symbol recommendation methods, including the online recommendation method and the offline recommendation method, meets the user demand in the offline and online states, then after the end of the current expression recommendation process, saves the expression symbol set by the user, the corresponding keyword, the geographic information, the time period and the like information as the scene record in the local, sends the scene record to the server when there is the network connection, saves it to the database, and trains the neural network model through the updated scene record, improves the accuracy of scene recognition, and further improves the accuracy of expression symbol recommendation.

[0137] The above specific embodiments further specifically describe the purposes, technical solutions and beneficial effects of the application. It should be understood that the above description is only for specific embodiments of the application and is not used to limit the protection scope of the application. It is particularly pointed out that any modification, equivalent replacement, improvement and the like made by those skilled in the art within the spirit and principle of the application should be included in the protection scope of the application.

Claims

1. A method for recommending an emoticon based on scene adaptation, characterized in that, The application comprises the following steps: determining the corresponding text keyword according to the input voice information; obtaining the current scene recognition result according to the detected several current scene identifiers and combining the preset neural network model; wherein the several first scene identifiers in the current area are obtained by detection, and the several first scene identifiers comprise SSID, GPS positioning and mobile network positioning; the several current scene identifiers are obtained by screening the several first scene identifiers through the preset scene identifier screening condition; the scene identifier screening condition comprises identifier text, set geographical area range and signal strength; matching the scene recognition result with the preset several scene records to obtain the first scene record corresponding to the scene recognition result; wherein the scene record comprises user-selected emoticons, corresponding keywords, geographical information and time period; obtaining the recommended emoticon list according to the text keyword, current scene recognition result and first scene record and combining the preset scene adaptation algorithm; wherein the recommended emoticon list is obtained by selecting in the local emoticon library according to the text keyword, current scene recognition result and first scene record through the preset scene adaptation algorithm, and the recommended emoticon list is displayed by using the preset sorting mode; the recommended emoticon list is dynamically updated in the recommended order of the emoticon list through the preset time threshold and the input voice corresponding text keyword obtained by the user terminal; the preset sorting mode comprises the use frequency of the emoticons in the scene record and the set emoticon recommendation threshold. 2.The method of claim 1, wherein, The training process of the preset neural network model comprises the following steps: receiving the new neural network model sent by the server in a timely manner, saving the new neural network model to the local of the user terminal, and deleting the old neural network model; the server trains the neural network model according to the several scene records saved in the server in a timely manner, obtains the new neural network model, and sends the new neural network model to the user terminal; the scene record comprises scene identifier, historical use emoticon, text keyword and time period. 3.The method of claim 1, wherein, The matching of the scene recognition result with the preset several scene records to obtain the corresponding first scene record comprises the following steps: determining the method of matching the scene record by the user terminal by judging whether the current user terminal is in network connection state; if the user terminal is in network connection state, sending the scene recognition result to the server, and receiving the first scene record matched with the scene recognition result obtained by the server according to the several scene records pre-stored in the server; if the user terminal is not in network connection state, obtaining the first scene record matched with the scene recognition result according to the several scene records saved in the user terminal. 4.The method of claim 1, wherein, The obtaining of the recommended emoticon list comprises the following steps: after obtaining the recommended emoticon list, the user terminal saves the user-selected emoticon, the text keyword corresponding to the voice, and the current scene recognition result as the scene record in the local. The user terminal sends the saved scene record to a server terminal through a network connection; The server terminal updates the scene record corresponding to the information in the scene record sent by the user terminal in the database in a timely manner. 5.The method of claim 1, wherein, Before determining the corresponding text keyword according to the input voice information, the method specifically comprises: Setting an interface according to the emoji information of the user terminal to set the emoji information; the emoji information comprises an emoji recommendation threshold and an emoji selection method; the emoji recommendation threshold comprises an emoji recommendation quantity and a recommendation priority level; the emoji selection method comprises clicking, shaking and gazing; The user terminal saves the set emoji information to the local terminal. 6.A scene-adaptive-based emoticon recommendation system, characterized by, The system comprises a text analysis module, a scene recognition module, a scene detection module and a recommendation module; The text analysis module is configured to determine the corresponding text keyword according to the input voice information; The scene recognition module is configured to obtain a current scene recognition result by combining a preset neural network model according to a plurality of detected current scene identifiers; the scene recognition module comprises a screening unit; the screening unit is configured to obtain a plurality of first scene identifiers in a current area by detection, wherein the plurality of first scene identifiers comprise an SSID, a GPS positioning and a mobile network positioning; and the plurality of first scene identifiers are screened by a preset scene identifier screening condition to obtain a plurality of current scene identifiers; the scene identifier screening condition comprises an identifier text, a set geographical area range and a signal strength; The scene detection module is configured to match the scene recognition result with a plurality of preset scene records to obtain a first scene record corresponding to the scene recognition result; the scene record comprises a user-selected emoji, a corresponding keyword, geographical information and a time period; The recommendation module is configured to obtain a recommended emoji list according to the text keyword, the current scene recognition result and the first scene record by combining a preset scene adaptation algorithm; the recommended emoji list is obtained by selecting in a local emoji library according to the text keyword, the current scene recognition result and the first scene record by the preset scene adaptation algorithm, and the recommended emoji list is displayed by using a preset sorting method; the recommended emoji list is dynamically updated in the recommended order of the emoji list by a preset time threshold and the text keyword corresponding to the input voice obtained by the user terminal in a continuous updating manner; the preset sorting method comprises a use frequency of the emoji in the scene record and a set emoji recommendation threshold.

7. The context-adaptive emoji recommendation system of claim 6, wherein, The scene recognition module further comprises a training unit; The training unit is configured to receive a new neural network model sent by the server terminal in a timely manner, save the new neural network model to the local terminal of the user terminal, and delete an old neural network model. The server side trains the neural network model according to a plurality of scene records stored in the server side, obtains a new neural network model, and sends the new neural network model to the user side; the scene record includes a scene identifier, a history of using emoticons, a text keyword, and a time period.

8. The context-adaptive emoji recommendation system of claim 6, wherein, The scene detection module includes a judgment unit and a detection unit; The judgment unit is configured to determine whether the current user side is in a network connection state and determine a method for matching the scene record by the user side; The detection unit is configured to, when the user side is in the network connection state, send the scene recognition result to the server side, receive a first scene record matched with the scene recognition result obtained by the server side according to a plurality of scene records pre-stored in the server side; and When the user side is not in the network connection state, a first scene record matched with the scene recognition result is obtained according to a plurality of scene records stored in the user side.

Images