WO2020024455A1 - 基于上下文的输入方法、装置、存储介质及计算机设备 - Google Patents

基于上下文的输入方法、装置、存储介质及计算机设备 Download PDF

Info

Publication number
WO2020024455A1
WO2020024455A1 PCT/CN2018/111681 CN2018111681W WO2020024455A1 WO 2020024455 A1 WO2020024455 A1 WO 2020024455A1 CN 2018111681 W CN2018111681 W CN 2018111681W WO 2020024455 A1 WO2020024455 A1 WO 2020024455A1
Authority
WO
WIPO (PCT)
Prior art keywords
information
target
dialog
dialogue
local user
Prior art date
Application number
PCT/CN2018/111681
Other languages
English (en)
French (fr)
Inventor
齐泽青
Original Assignee
平安科技(深圳)有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技(深圳)有限公司 filed Critical 平安科技(深圳)有限公司
Publication of WO2020024455A1 publication Critical patent/WO2020024455A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F3/00Input arrangements for transferring data to be processed into a form capable of being handled by the computer; Output arrangements for transferring data from processing unit to output unit, e.g. interface arrangements
    • G06F3/01Input arrangements or combined input and output arrangements for interaction between user and computer
    • G06F3/02Input arrangements using manually operated switches, e.g. using keyboards or dials
    • G06F3/023Arrangements for converting discrete items of information into a coded form, e.g. arrangements for interpreting keyboard generated codes as alphanumeric codes, operand codes or instruction codes
    • G06F3/0233Character input methods

Definitions

  • the present application relates to the field of communication technologies, and in particular, to a context-based input method, device, storage medium, and computer device.
  • An input method is an input editing tool. Users can use the input method to edit information, so as to achieve human-to-human or human-to-machine interaction.
  • the current input methods generally have a word association input function, which can predict the next word based on the user's last entered word and provide it to the user for selection, or obtain the corresponding word for the user to choose based on the pinyin acronym, or use fuzzy pronunciation association Match words to users.
  • the word prediction input function of the current input method is based on user input. Before the user does not enter any information or characters, the input method only counts high-frequency words based on historical statistics (such as "I", "you", “OK” "Etc.) and displayed for the user to choose, but the high-frequency vocabulary is not necessarily suitable for this input scenario. At this time, the user must actively input the corresponding content. The user's operation is tedious, time-consuming, labor-intensive, and low input efficiency. .
  • the present application provides a context-based input method, device, storage medium, and computer equipment, which are used to solve the defect of low efficiency of the existing input methods.
  • the target reply information is added to a text input field of the target dialog interface.
  • an embodiment of the present application further provides a context-based input device, including:
  • An acquisition module configured to acquire target dialog information sent by a non-local user in the target dialog interface when an information input instruction triggered by the local user on the target dialog interface is received;
  • a processing module configured to perform natural language understanding processing on the target dialog information, determine and display one or more reply information corresponding to the target dialog information according to a processing result, and display the processing result including the target dialog information An ordered sequence of corresponding feature words;
  • An input module is configured to add the target reply information to a text input field of the target dialog interface when the local user receives a selection instruction for the target reply information.
  • An embodiment of the present application further provides a non-volatile computer-readable storage medium. At least one computer-readable instruction is stored in the storage medium. When the computer-readable instruction is executed by a processor, the foregoing context-based input is implemented. method.
  • An embodiment of the present application further provides a computer device, including:
  • the memory is configured to store at least one computer-readable instruction, and the processor implements the above-mentioned context-based input method when executing the computer-readable instruction.
  • a context-based input method, device, storage medium, and computer device provided in the embodiments of the present application can automatically obtain target conversation information sent by a non-local user before a local user performs any input operation, and then determine a corresponding reply message.
  • the target reply message is added to the input field so that the local user can directly send the target reply message or edit the target reply message.
  • This method can automatically generate a target reply message before a local user performs any input operation, which simplifies user operations, saves operation time, can achieve quick responses, and improves user operation efficiency.
  • the dialogue information that is more important to the local user is selected based on the weight coefficient of the dialogue information, so as to facilitate subsequent automatic response to the important dialogue information.
  • Determining the weight coefficient of the dialogue information based on the multi-dimensional parameters can make the final weight coefficient more accurate.
  • the neural network model can quickly determine the response information, and at the same time call the third-party database under appropriate circumstances, can provide more complete and accurate response information; and based on the effective response information modified by the local user training the God will network model, can make the neural network The model is more in line with the needs of local users.
  • FIG. 1 is a flowchart of a context-based input method according to an embodiment of the present application
  • FIG. 2 is a flowchart of obtaining target dialog information in an embodiment of the present application
  • FIG. 3 is a flowchart of determining a reply message corresponding to the dialog message in the embodiment of the present application
  • FIG. 4 is a first structural diagram of a context-based input device in an embodiment of the present application.
  • FIG. 5 is a structural diagram of an acquisition module in an embodiment of the present application.
  • FIG. 6 is a structural diagram of a processing module in an embodiment of the present application.
  • FIG. 7 is a second structural diagram of a context-based input device according to an embodiment of the present application.
  • FIG. 8 is a schematic structural diagram of a terminal according to an embodiment of the present application.
  • An embodiment of the present application provides a context-based input method, whose execution subject may specifically be a terminal of a local user. As shown in FIG. 1, the method includes steps 101-103:
  • Step 101 When receiving an information input instruction triggered by a local user on the target dialog interface, obtain target dialog information sent by a non-local user in the target dialog interface.
  • the target dialogue interface is an interface for a local user to talk with other users, such as a chat interface of a chat tool; the local user can trigger the information input instruction in various ways, for example, the local user opens the target dialogue interface.
  • the information input instruction is automatically triggered at any time, or a local user operation (such as clicking, double-clicking, long-pressing, sliding, etc.) is used as an input field for inputting text in a target dialog interface.
  • a non-local user is one or more other users that are different from the local user. Both local and non-local users can submit information at a certain point in time. Each time a terminal acquires a user (including local and non-local users) For dialogue information, a point in time can be determined, that is, a piece of dialogue information corresponds to a point in time.
  • a local user when a local user needs to automatically reply to the dialog information sent by other users (that is, non-local users) at the current time point, an information input instruction is triggered on the target dialog interface, and then the target dialog information is determined.
  • the target time point may be the previous time point adjacent to the current time point, that is, the last dialog information sent by the non-local user in the target dialog interface is used as the target dialog information, and the dialog information obtained at this time is the latest message received by the terminal; At the same time, multiple conversation information submitted at multiple target time points can be acquired; or, which time point is determined as the target time point according to a preset rule.
  • each piece of information entered by the user corresponds to a point in time. If the local user is A and the other users chatting with the local user are B, user B sends a message d to user A at time point c. And user A has not responded to the message d for the time being, the target time point can be time point c, and the dialogue information submitted by user B is message d. If the previous time point of the current time point is the time point when the local user sent the message, it means that the local user may have responded to the message sent by other users. At this time, the method provided in the embodiment of this application can be executed, or the local user can continue to be found The second dialog information that has not been replied, and then the corresponding target time point is determined, and the subsequent process provided by the embodiment of this application is performed.
  • a target time point may be determined according to a user selection instruction input by a local user. For example, before the local user A responds to the message, three other users B, C, and D all send messages (or one user sends multiple messages, etc.). The user can press and hold (or click, double-click, etc.) a certain message. Message to the time point corresponding to the message as the target time point.
  • the message of the question sentence is determined based on the semantics of each message, and the message of the question sentence is automatically responded preferentially; when there are multiple messages of the question sentence, it is determined according to the semantics or the number of keywords Which interrogative message is more relevant to the content of the user's previous chat, and the most relevant interrogative message is given priority.
  • Step 102 Perform natural language understanding processing on the target dialog information, and determine and display one or more reply messages corresponding to the target dialog information according to the processing result; wherein the processing result includes an orderly feature word corresponding to the target dialog information. sequence.
  • the natural language understanding processing may include preprocessing such as segmentation and de-stopping.
  • Existing natural language understanding processing (NLU) technology may be used to perform natural language processing on the dialogue information, which may determine The semantics of the dialogue information.
  • NLU natural language understanding processing
  • the processing result of the natural language understanding processing of the target dialogue information can be determined based on the part-of-speech analysis and syntactic analysis of the words in the target dialogue information.
  • the processing result is An ordered sequence of feature words corresponding to the target dialogue information; wherein the feature words may be words determined after preprocessing the target dialogue information; the ordered sequence may be determined according to the order of the feature words in the target dialogue information, or according to a standard
  • the grammar generates an ordered sequence of feature words.
  • the response information corresponding to the dialogue information can be determined by searching the knowledge base or other methods (for example, using a neural network model).
  • the knowledge base stores the ordered sequence of feature words and the response information.
  • the mapping relationship between them, and the mapping relationship saved in the knowledge base can be updated in real time. For example, the target conversation information is "what do you want to eat tonight?", "What do you want to eat tonight?”, Or "what do you want to eat tonight?” Based on natural language understanding processing, it can be determined that the above three represent the same kind.
  • the feature words include three, which are “tonight”, “eat”, “what”, the ordered sequence of feature words can be “tonight-eat-what", or “Eat-what-tonight”; at this time, the response information corresponding to the ordered sequence of feature words can be determined according to the mapping relationship saved in the knowledge base; the response information can be one or more, such as “tonight Eat noodles “,” eat fruit tonight “,” eat barbecue tonight “, and so on; the reply information can be specifically determined based on the historical behavior of the local user, or it can be determined by calling the local user's schedule, memo, and so on.
  • each feature word in the processing result is provided with a feature attribute
  • the feature attribute includes a classification attribute and a grammatical attribute
  • the classification attribute indicates a category described in the current environment of the feature word, such as restaurant, location, music, etc.
  • the grammatical attribute indicates the part-of-speech (such as a noun, a verb, etc.) and the component (such as a subject, predicate, etc.) of the feature word in the target dialog information.
  • the classification attribute does not exist, it can be marked as null; the classification attribute is used to distinguish the meaning of the feature word. For example, “What do you think of Jiangnan?"
  • the characteristic word "Jiangnan” can represent a geographic location or a song. At this time, it can't be distinguished only by the grammatical attribute, but by the classification attribute "location” or "music "To distinguish them.
  • Step 103 When the local user's selection instruction for the target reply information is received, add the target reply information to the text input field of the target dialog interface.
  • the selection instruction may be an automatically generated instruction or an automatically triggered instruction; for example, after determining the reply information in step 102, directly selecting the reply information as the target reply information (equivalent to automatically generating a selection instruction, The user cannot perceive it); or after the reply message is determined, the selection instruction is triggered based on a certain trigger condition (for example, the user does not operate for a long time, or the terminal receives information from other users, etc.).
  • the selection instruction may be an instruction actively input by a local user.
  • the reply message when the reply message is displayed in step 102, the reply message may be displayed in the text input field of the target dialog interface, or may be displayed around the text input field or at another preset position, as long as it can be seen by the local user. Just fine. When there is only one reply message, it can be displayed directly; when there are multiple reply messages, multiple reply messages can be displayed side by side (or side by side); or after multiple reply messages are sorted, the reply messages are displayed in order.
  • displaying a reply message may require a triggering instruction; for example, the target reply message is only displayed when a local user clicks on a text entry field.
  • the local user can select a reply message as the target reply message by operating the terminal. Specifically, the user can select the reply message by clicking, double-clicking, long-pressing, or sliding the reply message; and sending a message to the local user Previously, the reply message was always displayed, that is, the local user can select the desired reply message at any time. At the same time, instead of sending immediately to avoid user selection errors, it is also possible to make the local user edit the reply message in the text input field and then send the edited reply message afterwards.
  • the embodiment of the present application provides a context-based input method. Before a local user performs any input operation, the target conversation information sent by a non-local user can be automatically obtained, and then the corresponding reply message is determined and the target reply message is added. To the input field, it is convenient for local users to directly send the target reply message or edit the target reply message. This method can automatically generate a target reply message before a local user performs any input operation, which simplifies user operations, saves operation time, can achieve quick responses, and improves user operation efficiency.
  • step 101 Another embodiment of the present application provides a context-based input method.
  • the method includes steps 101-103 in the foregoing embodiment.
  • steps 101-103 For implementation principles and technical effects, refer to the embodiment corresponding to FIG. 1.
  • “obtaining target dialog information sent by a non-local user in the target dialog interface” in step 101 includes steps 1011-1013:
  • Step 1011 Use the time point of the latest submission by the local user as the starting time point.
  • the information received before the local user once sent the information is defaulted to the information that has been replied, that is, the information received before is not automatically replied.
  • the time point when the local user last sent information that is, the start time point is determined, and the start time point is used as a division point.
  • Step 1012 Determine all the conversation information sent by the non-local user between the start time point and the current time point, and determine the information parameters corresponding to each conversation information.
  • the information parameters include the information content, the acquisition time, and other user's information. User ID.
  • the time point of the dialog information may be directly used as the target time point.
  • other users may send multiple messages (that is, conversation information), or multiple other users have sent conversation information during the group chat.
  • the starting time point and the current time point There are multiple messages between each other.
  • the dialogue information is used to determine which dialogue information is the information that needs to be automatically answered according to the weight coefficient of the dialogue information. That is, when there are multiple dialogue information between the start time point and the current time point, the step is performed. 1011-1013.
  • the information parameters of each piece of dialogue information are determined, and the information content is the text content contained in the dialogue information, and the acquisition time is the local terminal.
  • the user ID is used to distinguish different users, and one user corresponds to a unique user ID.
  • Step 1013 Determine the weight coefficient of the corresponding dialogue information according to the information parameter, and use the dialogue information with the largest weight coefficient or the weight coefficient greater than a preset threshold as the target dialogue information, and the time point corresponding to the target dialogue information is the target time point.
  • the weight coefficient of the dialogue information indicates the importance of the dialogue information to the local user. The larger the weight coefficient, the more important the dialogue information is to the local user. After determining the weight coefficients of all the dialogue information, use the dialogue information with the largest weight coefficient as the final target dialogue information, or use the dialogue information with a weight coefficient greater than a preset threshold as the final target dialogue information. Situation or local user settings. Based on the weight coefficient of the dialogue information, the dialogue information that is more important to the local user is selected, thereby facilitating subsequent automatic reply to the important dialogue information.
  • the weight coefficient may be determined based on various methods.
  • the above step 1013 "determining the weight coefficient of the corresponding dialog information according to the information parameter" may include steps A1-A5:
  • Step A1 Use one piece of dialogue information obtained between the start time point and the current time point as the first dialogue information.
  • each dialogue information between the start time point and the current time point is determined in accordance with steps A1-A5.
  • the first dialogue information is only one of the dialogue information, and different dialogues may be separately used.
  • the information is used as the first dialog information, and then the corresponding weight coefficients are determined according to steps A2-A5.
  • Step A2 Determine the similarity between the information content of the first dialogue information and the information content of the second dialogue information, and determine the similarity coefficient of the first dialogue information according to all the similarities; the second dialogue information is the first dialogue information divided Outside the second conversation information.
  • the similarity coefficient of the first dialog information is first determined according to the information content of the first dialog information. Specifically, it is assumed that there are n pieces of dialogue information between the start time point and the current time point, c i represents the information content of the i-th piece of dialogue information, and i ⁇ [1, n], that is, n represents the start time point
  • the number of dialog messages from the current point in time; the similarity between the information content of the i and j dialog messages is s ij , j ⁇ [1, n], and i ⁇ j; According to the ratio of the same number of characters to the total number of characters between the two pieces of information, the similarity between the two can be determined in other ways.
  • the similarity between two identical dialog messages is 1.
  • the i-th conversation information is the first conversation information.
  • the average value of the similarity between the information content of the first conversation information and the information content of all other second conversation information may be used as the first conversation information.
  • Similarity coefficient S i , ie Where i ⁇ j. The larger the similarity coefficient, the higher the correlation between the first dialog information and other dialog information (that is, the second dialog information), and a reply can be given priority.
  • Step A3 Determine the time difference between the acquisition time of the first dialog information and the current time point, and determine the receiving order of the first dialog information according to the acquisition time of the first dialog information.
  • the time difference indicates the length of time during which the first conversation information is obtained.
  • the reception order indicates that the first conversation information is in all The conversation information is the first one obtained by the local terminal. The smaller the receiving order, the earlier the acquisition is. In general, the response should be given priority. It should be noted that some users are more accustomed to replying to the information they just received. At this time, the time difference is small or the receiving order is greater, the priority is to reply, which can be determined according to the habits or settings of the local user.
  • Step A4 Determine the intimacy between the corresponding non-local user and the local user according to the user identification of the first conversation information, and determine that the user identification of the first conversation information submits the conversation information between the start time point and the current time point. The number of submissions.
  • whether to reply to this piece of information is determined preferentially according to the intimacy between the local user and other users corresponding to the first conversation information.
  • the intimacy between the local user and the other user can be determined according to the chat history, chat frequency, etc. of the other user; the local user can also set the intimacy of one other user to himself.
  • the greater the amount of conversation information submitted by the other user between the start time point and the current time point the more it indicates that the other user needs to be preferentially responded.
  • Step A5 Determine the weighting coefficient of the first dialog information according to the similarity coefficient, time difference, receiving order, intimacy, and number of submissions; the weight coefficient is positively related to the similarity coefficient, intimacy, and number of submissions, and the weight coefficient is related to time difference and receiving
  • the reciprocal of the order is positive or negative.
  • a positive correlation between the weight coefficient and the similarity coefficient indicates that the larger the similarity coefficient, the larger the weight coefficient; at the same time, the relationship between the weight coefficient and the time difference and the receiving order needs to be based on the actual situation of the local user It depends.
  • the weight coefficients of the dialog information are jointly determined based on multi-dimensional parameters such as the similarity coefficient, the time difference, the receiving order, the intimacy, and the number of submissions, which can make the finally determined weight coefficients more accurate.
  • step 102 "determining one or more reply information corresponding to the target dialog information according to the processing result" includes steps 1021-1024:
  • Step 1021 Input the processing result of the target dialog information into a preset information matching model, and use the output result of the information matching model as temporary response information corresponding to the target dialog information;
  • the information matching model is a plurality of natural language understanding processing
  • the target dialogue information and the corresponding response information are used as samples to input the neural network model and trained.
  • a neural network model is preset, a plurality of target dialog information processed by natural language understanding is used as an input, and a user's response to the target dialog information is used as an output to train a neural network model, and a neural network model is obtained.
  • the specific parameters of the neural network model after training as the information matching model. After obtaining the processing result of the target dialog information that the local user needs to reply to, use the processing result as the input of the information matching model to obtain the corresponding information for replying to the processing result, that is, the temporary reply information.
  • Step 1022 Determine whether there is a call identifier for calling the third-party database in the temporary response information, and when the call identifier does not exist in the temporary response information, use the temporary response information as the response information corresponding to the dialog information.
  • Step 1023 When there is a call identifier in the temporary response information, query a third-party database corresponding to the call identifier, and supplement the temporary response information according to the query result, and use the supplementary temporary response information as the response information corresponding to the target dialog information.
  • a third-party database may also be used when determining the reply information.
  • the third-party database is a database that needs to be updated in real time, and may specifically include a weather database, a road condition database, a flight database, a local database, and the like; Data related to local users, such as local users' schedules, memos, etc.
  • the call identifier is used to indicate whether a third-party database needs to be called. When the third-party database does not need to be called, the call identifier may be null; when the third-party database needs to be called, the call identifier is consistent with the required third-party database; A reply message may not include multiple call identifiers.
  • the conversation information submitted by other users is "How is your weather?"
  • the reply information can be directly determined according to the neural network model, there is no need to call a third-party database.
  • the embodiment of the present application can quickly determine the reply information through a neural network model, and at the same time call a third-party database under appropriate conditions, which can provide more complete and accurate reply information.
  • step 103 add the target reply information to the text input field of the target dialog interface
  • the method further includes a subsequent process, which specifically includes steps B1-B2:
  • Step B1 When receiving the modification instruction input by the local user, modify the response information according to the modification instruction, and send the modified response information as valid response information.
  • Step B2 Take the target dialog information as input and valid response information as output, train and update the preset information matching model.
  • the information matching model is a model trained by inputting a plurality of target dialog information and corresponding response information processed by natural language understanding into a neural network model as a sample.
  • the local user can directly send the target reply message, or edit the target reply message.
  • the local user can enter a modification instruction to edit the target reply message.
  • the modified response information (that is, valid response information) of the local user may be used as a sample.
  • the embodiment of the present application provides a context-based input method.
  • the target conversation information sent by a non-local user can be automatically obtained, and then the corresponding reply message is determined and the target reply message is added.
  • This method can automatically generate a target reply message before a local user performs any input operation, which simplifies user operations, saves operation time, can achieve quick responses, and improves user operation efficiency.
  • the dialogue information that is more important to the local user is selected based on the weighting coefficient of the dialogue information, so that it is convenient to automatically respond to this important dialogue information in the future.
  • Determining the weight coefficient of the dialogue information based on the multi-dimensional parameters can make the final weight coefficient more accurate.
  • the neural network model can quickly determine the response information, and at the same time call the third-party database under appropriate circumstances, can provide more complete and accurate response information; and based on the effective response information modified by the local user training the God will network model, can make the neural network The model is more in line with the needs of local users.
  • An embodiment of the present application provides a context-based input device, as shown in FIG. 4, including:
  • An obtaining module 41 is configured to obtain target dialog information sent by a non-local user in a target dialog interface when an information input instruction triggered by a local user on the target dialog interface is received;
  • a processing module 42 is configured to perform natural language understanding processing on the target dialog information, determine and display one or more reply information corresponding to the target dialog information according to a processing result, and display the processing result including a feature corresponding to the target dialog information An ordered sequence of words;
  • the input module 43 is configured to add target reply information to a text input field of a target dialog interface when a local user's selection instruction for the target reply information is received.
  • the obtaining module 41 includes:
  • a time determining unit 411 configured to use the time point of the latest submission information of the local user as the starting time point;
  • the obtaining unit 412 is configured to determine all the conversation information sent by the non-local user between the start time point and the current time point, and determine information parameters corresponding to each conversation information.
  • the information parameters include information content, acquisition time, and User ID of other users;
  • the processing unit 413 is configured to determine a weight coefficient of the corresponding dialogue information according to the information parameter, and use the dialogue information with the largest weight coefficient or a weight coefficient greater than a preset threshold as the target dialogue information, and the time point corresponding to the target dialogue information is the target time point.
  • the processing unit 413 includes:
  • a preprocessing subunit configured to use a piece of dialog information acquired between a start time point and a current time point as first dialog information
  • a first determining subunit configured to determine a similarity between the information content of the first dialog information and the information content of the second dialog information, and determine a similarity coefficient of the first dialog information according to all the similarities;
  • the second dialog information is Second dialog information other than the first dialog information;
  • a second determining subunit configured to determine a time difference between an acquisition time of the first dialog information and a current time point, and determine a receiving order of the first dialog information according to the acquisition time of the first dialog information
  • a third determining subunit configured to determine the intimacy between the corresponding non-local user and the local user according to the user identifier of the first dialog information, and determine the user identifier of the first dialog information at the start time point and the current time point The number of submissions of dialogue information between submissions;
  • a processing subunit configured to determine a weight coefficient of the first dialog information according to the similarity coefficient, time difference, receiving order, intimacy, and number of submissions; the weight coefficient has a positive correlation with the similarity coefficient, intimacy, and number of submissions, There is a positive or negative correlation between the time difference and the reciprocal of the receiving order.
  • the processing module 42 includes:
  • An information determining unit 422 is configured to use an output result of the information matching model as temporary response information corresponding to the target dialog information;
  • the information matching model is a plurality of target dialog information and corresponding responses processed by natural language understanding.
  • the information is input to the neural network model as a sample to train the model;
  • the processing unit 423 is configured to determine whether there is a call identifier for calling a third-party database in the temporary response information, and the third-party database includes one or more of a weather database, a road condition database, a flight database, and a local database; When the caller ID does not exist, the temporary reply information is used as the reply information corresponding to the target dialog information; when the caller ID exists in the temporary reply message, the third-party database corresponding to the caller ID is queried, and the temporary reply information is supplemented according to the query result , Use the supplementary temporary reply information as the reply information corresponding to the target conversation information.
  • the apparatus further includes a modification module 44 and a training module 45;
  • the modification module 44 is used to modify the target response information according to the modification instruction when receiving the modification instruction input by the local user, and use the modified target response information as a valid response.
  • the training module 45 is configured to take the target dialog information as input and the effective response information as output, and train and update a preset information matching model.
  • the information matching model is a plurality of target dialog information and corresponding response information processed by natural language understanding as samples. The model trained after inputting the neural network model.
  • the obtaining module 41 is specifically configured to use the last piece of dialog information sent by the non-local user in the target dialog interface as the target dialog information.
  • An embodiment of the present application further provides a non-volatile computer-readable storage medium.
  • the non-volatile computer-readable storage medium stores computer-readable instructions, which includes a program for executing the foregoing context-based input method.
  • the computer-readable instructions may execute the method in any of the method embodiments described above.
  • the non-transitory computer-readable storage medium may be any available medium or data storage device that can be accessed by a computer, including but not limited to magnetic storage (such as a floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.), Optical memory (such as CD, DVD, BD, HVD, etc.), and semiconductor memory (such as ROM, EPROM, EEPROM, non-volatile memory (NAND, FLASH), solid-state hard disk (SSD), etc.).
  • magnetic storage such as a floppy disk, hard disk, magnetic tape, magneto-optical disk (MO), etc.
  • Optical memory such as CD, DVD, BD, HVD, etc.
  • semiconductor memory such as ROM, EPROM, EEPROM, non-volatile memory (NAND, FLASH), solid-state hard disk (SSD), etc.
  • FIG. 8 shows a structural block diagram of a computer device according to another embodiment of the present application.
  • the computer device 1100 may be a host server with a computing capability, a personal computer PC, or a portable portable computer or computer device.
  • the specific embodiments of the present application do not limit the specific implementation of the computer equipment.
  • the computer device 1100 includes at least one processor 1110, communications interface 1120, memory 1130, and a bus 1140. Among them, the processor 1110, the communication interface 1120, and the memory 1130 complete communication with each other through the bus 1140.
  • the communication interface 1120 is configured to communicate with a network element, where the network element includes, for example, a virtual machine management center, shared storage, and the like.
  • the processor 1110 is configured to execute computer-readable instructions.
  • the processor 1110 may be a central processing unit CPU, or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.
  • ASIC application specific integrated circuit
  • the memory 1130 is used for executable instructions.
  • the memory 1130 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), for example, at least one disk memory.
  • the memory 1130 may also be a memory array.
  • the memory 1130 may also be divided into blocks, and the blocks may be combined into a virtual volume according to a certain rule.
  • the computer-readable instructions stored in the memory 1130 can be executed by the processor 1110, so that the processor 1110 can execute the method in any of the foregoing method embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Machine Translation (AREA)

Abstract

一种基于上下文的输入方法、装置、存储介质及计算机设备,其中,该方法包括:当接收到本地用户在目标对话界面触发的信息输入指令时,获取目标对话界面中非本地用户发送的目标对话信息(101);对目标对话信息进行自然语言理解处理,根据处理结果确定与目标对话信息相对应的一个或多个回复信息并展示(102);当接收到本地用户针对目标回复信息的选择指令时,将目标回复信息添加至目标对话界面的文本输入栏中(103)。该方法在本地用户未执行任何输入操作之前即可自动生成回复消息,简化了用户操作,节省了操作时间,可以实现快速回复,提高了用户操作效率。

Description

基于上下文的输入方法、装置、存储介质及计算机设备
本申请要求于2018年8月1日提交中国专利局、申请号为2018108641740、申请名称为“一种基于上下文的输入方法、装置、存储介质及终端”的中国专利申请的优先权,其全部内容通过引用结合在申请中。
技术领域
本申请涉及通信技术领域,特别涉及一种基于上下文的输入方法、装置、存储介质及计算机设备。
背景技术
输入法是一种输入编辑工具,用户可以利用输入法进行信息的编辑,从而实现人与人或者人与机器之间的交互。目前的输入法一般均设有词语联想输入功能,可以根据用户最后输入的词语预测下一个词并提供给用户选择,或者根据拼音首字母缩写获取对应的词语提供给用户选择,或者采用模糊音联想匹配词语给用户选择。而目前输入法的词语联想输入功能均是基于用户输入的,在用户未输入任何信息或字符之前,输入法只是根据历史统计方式统计出高频词汇(例如“我”,“你”,“好”等)并显示出来以供用户选择,但是该高频词汇并不一定适用于本次的输入场景,此时需要用户主动输入相应内容才可以,用户操作繁琐,且费时费力、输入效率较低。
发明内容
本申请提供一种基于上下文的输入方法、装置、存储介质及计算机设备,用以解决现有输入方式效率低的缺陷。
本申请实施例提供的一种基于上下文的输入方法,包括:
当接收到本地用户在目标对话界面触发的信息输入指令时,获取所述目标对话界面中非本地用户发送的目标对话信息;
对所述目标对话信息进行自然语言理解处理,根据处理结果确定与所述目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
当接收到所述本地用户针对目标回复信息的选择指令时,将所述目标回复信息添加至所述目标对话界面的文本输入栏中。
基于同样的发明构思,本申请实施例还提供一种基于上下文的输入装置,包括:
获取模块,用于当接收到本地用户在目标对话界面触发的信息输入指令时,获取所述目标对话界面中非本地用户发送的目标对话信息;
处理模块,用于对所述目标对话信息进行自然语言理解处理,根据处理结果确定与所述目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
输入模块,用于当接收到所述本地用户针对目标回复信息的选择指令时,将所述目标回复信息添加至所述目标对话界面的文本输入栏中。
本申请实施例还提供一种非易失性计算机可读存储介质,所述存储介质中存储有至少一计算机可读指令,所述计算机可读指令被处理器执行时实现上述的基于上下文的输入方法。
本申请实施例还提供一种计算机设备,包括:
处理器、存储器、通信接口和通信总线,所述处理器、所述存储器和所述通信接口通过所述通信总线完成相互间的通信;
所述存储器用于存放至少一计算机可读指令,所述所述处理器执行所述计算机可读指令时实现上述的基于上下文的输入方法。
本申请实施例提供的一种基于上下文的输入方法、装置、存储介质及计算机设备,在本地用户未执行任何输入操作之前即可自动获取非本地用户发送的目标对话信息,进而确定相应的回复消息并将其中的目标回复消息添加至输入栏,方便本地用户直接发送该目标回复消息或编辑目标回复消息。该方法在本地用户未执行任何输入操作之前即可自动生成目标回复消息,简化了用户操作,节省了操作时间,可以实现快速回复,提高了用户操作效率。基于对话信息的权重系数来选定对本地用户比较重要的对话信息,从而方便后续优先对该重要的对话信息进行自动回复。基于多维参数共同确定对话信息的权重系数,可以使得最终确定的权重系数更加精确。通过神经网络模型可以快速确定回复信息,同时在适当的情况下调用第三方数据库,能够提供更加完善精确的回复信息;且基于本地用户修改后的有效回复信息训练神将网络模型,可以使得神经网络模型更加符合本地用户的需求。
本申请的其它特征和优点将在随后的说明书中阐述,并且,部分地从说明书中变得显而易见,或者通过实施本申请而了解。本申请的目的和其他优点可通过在所写的说明书、权利要求书、以及附图中所特别指出的结构来实现和获得。
下面通过附图和实施例,对本申请的技术方案做进一步的详细描述。
附图说明
附图用来提供对本申请的进一步理解,并且构成说明书的一部分,与本申请的实施例一起用于解释本申请,并不构成对本申请的限制。在附图中:
图1为本申请实施例中基于上下文的输入方法的流程图;
图2为本申请实施例中获取目标对话信息的流程图;
图3为本申请实施例中确定与对话信息相对应的回复信息的流程图;
图4为本申请实施例中基于上下文的输入装置的第一结构图;
图5为本申请实施例中获取模块的结构图;
图6为本申请实施例中处理模块的结构图;
图7为本申请实施例中基于上下文的输入装置的第二结构图;
图8为本申请实施例中终端的结构示意图。
具体实施方式
以下结合附图对本申请的优选实施例进行说明,应当理解,此处所描述的优选实施例仅用于说明和解释本申请,并不用于限定本申请。
本申请实施例提供的一种基于上下文的输入方法,其执行主体具体可以为本地用户的终端,参见图1所示,该方法包括步骤101-103:
步骤101:当接收到本地用户在目标对话界面触发的信息输入指令时,获取目标对话界面中非本地用户发送的目标对话信息。
本申请实施例中,目标对话界面为本地用户与其他用户进行对话的界面,例如某聊天工具的聊天界面等;本地用户可通过多种方式触发该信息输入指令,例如本地用户打开该目标对话界面时自动触发该信息输入指令、或者本地用户操作(例如单击、双击、长按、滑动等操作)目标对话界面中用于输入文本的输入栏等。非本地用户为与本地用户不同的一个或多个其他用户,本地用户和非本地用户均可以在某一时间点提交信息,终端每获取到一个用户(包括本地的用户和非本地用户)提交的对话信息,则可以确定一个时间点,即一条对话信息对应一个时间点。
本申请实施例中,当本地用户需要在当前时间点自动回复其他用户(即非本地用户)发送的对话信息时,在目标对话界面触发一个信息输入指令,进而确定目标对话信息,该目标对话信息为非本地用户发送的一个对话信息。具体的,可以首先确定一个目标时间点,该目标时间点即为该目标对话信息对应的时间点。目标时间点可以是与当前时间点相邻的上一个时间点,即将目标对话界面内最后一条非本地用户发送的对话信息作为目标对话信 息,此时获取的对话信息是终端最新接收到的消息;同时,可以获取多个目标时间点提交的多个对话信息;或者,根据预设的规则来确定哪一时间点为目标时间点。
具体的,在聊天场景下,用户输入的每条信息对应一个时间点,若本地用户为A,与本地用户聊天的其他用户为B,用户B在时间点c给用户A发送了一条消息d,且用户A暂时并未回复该消息d,则目标时间点可以是时间点c,用户B提交的对话信息即为消息d。若当前时间点的上一个时间点是本地用户自己发送消息的时间点,说明本地用户可能已经回复了其他用户发送的消息,此时可以执行本申请实施例提供的方法,也可以继续寻找本地用户还未回复的第二对话信息,进而确定相对应的目标时间点,并执行本申请实施例提供的后续流程。
可选的,当一个用户连续发送多条消息、或者处于群聊环境时,可以根据本地用户输入的选择用户指令来确定目标时间点。例如,在本地用户A回复消息之前,有三个其他用户B、C、D均发了消息(或者一个用户发送了多条消息等),用户可以通过长按(或单击、双击等)某一条消息来将该消息对应的时间点作为目标时间点。或者,当一个用户连续发送多条消息时,基于每条消息的语义确定疑问句式的消息,优先自动回复疑问句式的消息;在存在多个疑问句式的消息时,根据语义或者关键词数量来确定哪一个疑问句式的消息与用户之前聊天的内容更加相关,优先回复最相关的疑问句式消息。此外,在群聊时,可以同时自动回复一个用户发送的多条消息。例如,若其他用户C连续发送了多个消息,此时可以同时自动回复用户C发送的多个消息,即本地用户A可以以一条消息的形式回复用户C发送的多条消息。
步骤102:对目标对话信息进行自然语言理解处理,根据处理结果确定与目标对话信息相对应的一个或多个回复信息并展示;其中,该处理结果包括与目标对话信息对应的特征词的有序序列。
本申请实施例中,自然语言理解处理可以包含分词、去停用词等预处理过程;可以采用现有的自然语言理解处理(Natural Language Understanding,NLU)技术对对话信息进行自然语言处理,可以确定对话信息的语义;具体的,对目标对话信息进行预处理后,基于对目标对话信息中词语的词性分析以及句法分析可以确定该目标对话信息的自然语言理解处理的处理结果,该处理结果为与该目标对话信息对应的特征词的有序序列;其中,特征词可以是对目标对话信息进行预处理后确定的词语;可以根据目标对话信息中特征词的顺序确定有序序列,也可以根据标准语法生成特征词的有序序列。确定处理结果之后,通过检索知识库的方式或者其他方式(例如利用神经网络模型)等可以确定与该对话信息相对应的回复信息,该知识库中保存有特征词的有序序列与回复信息之间的映射关系,且可实 时更新该知识库中保存的映射关系。例如,目标对话信息为“今晚想吃什么?”、“晚上打算吃什么?”、或“今晚有什么想吃的?”,基于自然语言理解处理可以确定上述三者表示的是同一种含义“今晚吃什么?”,特征词包括三个,分别是“今晚”、“吃”、“什么”,特征词的有序序列可以为“今晚—吃—什么”,也可为“吃—什么—今晚”;此时根据知识库中保存的映射关系即可以确定特征词的有序序列所对应的回复信息;该回复信息可以有一个,也可以多个,例如“今晚吃面条”、“今晚吃水果”、“今晚吃烤肉”等;该回复信息具体可以基于本地用户的历史行为来确定,也可通过调用本地用户日程表、备忘录等的方式来确定。
可选的,处理结果中每个特征词设有特征属性,该特征属性包括分类属性和语法属性,其中,分类属性表示该特征词在当前环境下所述的类别,比如餐厅、位置、音乐等;语法属性表示该特征词在该目标对话信息中的词性(比如名词、动词等)和所属成分(比如主语、谓语等)。如上例中的有序序列“今晚—吃—什么”,其中特征词“今晚”的语法属性可以为“副词-时间状语”,“吃”的语法属性为“动词-谓语”。同时,若分类属性不存在,可以标记为空(null);分类属性用于对特征词的含义进行区分。例如,“你觉得江南怎么样?”其中的特征词“江南”可以表示地理位置,也可以表示一首歌,此时只单靠语法属性并不能区分,而通过分类属性“位置”或“音乐”可以对其进行区分。
步骤103:当接收到本地用户针对目标回复信息的选择指令时,将目标回复信息添加至目标对话界面的文本输入栏中。
本申请实施例中,该选择指令可以为自动生成的指令或自动触发的指令;例如,在步骤102确定回复信息之后,直接选定该回复信息作为目标回复信息(相当于自动生成一个选择指令,用户感知不到);或者在确定回复信息之后基于某种触发条件(例如用户长时间不操作、或者终端又接收到其他用户发来的信息等)来触发选择指令。
或者,该选择指令也可以为本地用户主动输入的指令。具体的,在步骤102中显示该回复信息时,该回复信息可以显示在目标对话界面的文本输入栏中,也可以显示在文本输入栏周围或其他的预设位置,只要可以使得本地用户看到即可。当只有一个回复信息时,直接显示即可;当有多个回复信息时,可以并排(或并列)显示多个回复信息;或者对多个回复信息排序后再按照顺序依次显示回复信息。同时,显示回复消息可以需要一个触发指令;例如,当本地用户点击文本输入栏时才会显示该目标回复消息。之后本地用户通过操作终端即可选择一个回复消息作为目标回复消息;具体的,用户可以通过单击、双击、长按或滑动回复消息的方式来选择该回复消息;且在本地用户发送某一消息之前,该回复消息一直处于显示状态,即本地用户可以随时选择所需的回复消息。同时,不立刻发送以 避免用户选择错误,还可以使得本地用户在文本输入栏中编辑该回复消息,之后发送编辑后的回复消息。
本申请实施例提供的一种基于上下文的输入方法,在本地用户未执行任何输入操作之前即可自动获取非本地用户发送的目标对话信息,进而确定相应的回复消息并将其中的目标回复消息添加至输入栏,方便本地用户直接发送该目标回复消息或编辑目标回复消息。该方法在本地用户未执行任何输入操作之前即可自动生成目标回复消息,简化了用户操作,节省了操作时间,可以实现快速回复,提高了用户操作效率。
本申请另一实施例提供一种基于上下文的输入方法,该方法包括上述实施例中的步骤101-103,其实现原理以及技术效果参见图1对应的实施例。同时,本申请实施例中,参见图2所示,步骤101中“获取目标对话界面中非本地用户发送的目标对话信息”包括步骤1011-1013:
步骤1011:将本地用户最新提交信息的时间点作为起始时间点。
本申请实施例中,为了提高处理效率,将本地用户曾经发送信息之前所接收到的信息均默认为已经回复了的信息,即不自动回复之前所接收到的信息。具体的,确定本地用户最后一次发送信息的时间点,即起始时间点,将该起始时间点作为一个分割点。
步骤1012:确定在起始时间点与当前时间点之间所接收的非本地用户发送的所有对话信息,并确定每条对话信息对应的信息参数,信息参数包括信息内容、获取时间和其他用户的用户标识。
本申请实施例中,若在起始时间点与当前时间点之间仅存在一条消息,则可以直接将该对话信息的时间点作为目标时间点。此外,在起始时间点之后,可能其他用户会发送多条消息(即对话信息),或者在群聊时多个其他用户均发送了对话信息,则此时在起始时间点与当前时间点之间存在多条消息,此时根据对话信息的权重系数确定哪条对话信息才是需要自动回复的信息;即当起始时间点与当前时间点之间存在多条对话信息时,才执行步骤1011-1013。
具体的,当起始时间点与当前时间点之间存在多条对话信息时,确定每条对话信息的信息参数,其中的信息内容即为该对话信息所含的文本内容,获取时间为本地终端获取到该对话信息的时间,用户标识用于区分不同的用户,一个用户对应一个唯一的用户标识。
步骤1013:根据信息参数确定相应对话信息的权重系数,将权重系数最大或权重系数大于预设阈值的对话信息作为目标对话信息,目标对话信息对应的时间点为目标时间点。
本申请实施例中,对话信息的权重系数表示该对话信息对本地用户的重要程度,权重系数越大,说明该对话信息对本地用户越重要。在确定所有对话信息的权重系数后,将权 重系数最大的对话信息作为最终确定的目标对话信息,或者将权重系数大于某个预设阈值的对话信息作为最终确定的目标对话信息,具体可以根据实际情况或本地用户的设置而定。基于对话信息的权重系数来选定对本地用户比较重要的对话信息,从而方便后续对该重要的对话信息进行自动回复。
本申请实施例中可以基于多种方式来确定权重系数,可选的,上述步骤1013“根据信息参数确定相应对话信息的权重系数”可以包括步骤A1-A5:
步骤A1:将在起始时间点与当前时间点之间所获取的一个对话信息作为第一对话信息。
本申请实施例中,在起始时间点与当前时间点之间的每个对话信息均按照步骤A1-A5来确定权重系数,第一对话信息只是其中的一个对话信息,可以分别将不同的对话信息作为第一对话信息,进而根据步骤A2-A5来确定相应的权重系数。
步骤A2:确定第一对话信息的信息内容与第二对话信息的信息内容之间的相似度,根据所有的相似度确定第一对话信息的相似度系数;第二对话信息为除第一对话信息以外的第二对话信息。
本申请实施例中,首先根据第一对话信息的信息内容来确定第一对话信息的相似度系数。具体的,设在起始时间点与当前时间点之间共n条对话信息,c i表示第i条对话信息的信息内容,且i∈[1,n],即n表示在起始时间点与当前时间点之间对话信息的条数;第i条与第j条对话信息的信息内容之间的相似度为s ij,j∈[1,n],且i≠j;其中,具体可以根据两个信息内容之间相同字符数占总字符数的比值作为二者之间的相似度,也可以其他方式确定相似度,两个完全相同的对话信息之间的相似度为1。此时,假设第i条对话信息为第一对话信息,此时可以将第一对话信息的信息内容与其他所有的第二对话信息的信息内容之间的相似度的平均值作为第一对话信息的相似度系数S i,即:
Figure PCTCN2018111681-appb-000001
其中i≠j。相似度系数越大,说明第一对话信息与其他的对话信息(即第二对话信息)之间的相关度越高,可以优先回复。
步骤A3:确定第一对话信息的获取时间与当前时间点之间的时间差,并根据第一对话信息的获取时间确定第一对话信息的接收顺位。
本申请实施例中,该时间差表示获取到第一对话信息的时长,时间差越大,表示时长越长,一般情况下更应该优先回复;同样的,接收顺位表示该第一对话信息在所有的对话信息中是第几个被本地终端获取的,接收顺位越小,表示获取的越早,一般情况下应该优先回复。需要说明的是,部分用户更加习惯回复刚收到的信息,此时时间差小的或者接收 顺位大的则更优先回复,具体可以根据本地用户的习惯或设置而定。
步骤A4:根据第一对话信息的用户标识确定相对应的非本地用户与本地用户之间的亲密度,并确定第一对话信息的用户标识在起始时间点与当前时间点之间提交对话信息的提交数量。
本申请实施例中,根据本地用户与该第一对话信息对应的其他用户之间的亲密度来确定是否优先回复该条信息。具体的,可以根据本地用户与该其他用户之间的聊天记录、聊天频率等来确定二者之间的亲密度;本地用户也可自行设置某个其他用户对自己的亲密度。同时,该其他用户在起始时间点与当前时间点之间提交对话信息的数量越多,越说明该其他用户更需要被优先回复。
步骤A5:根据相似度系数、时间差、接收顺位、亲密度和提交数量确定第一对话信息的权重系数;权重系数与相似度系数、亲密度和提交数量呈正相关关系,权重系数与时间差和接收顺位的倒数呈正相关关系或负相关关系。
本申请实施例中,权重系数与相似度系数之间呈正相关关系表示相似度系数越大,权重系数越大;同时,权重系数与时间差和接收顺位之间的关系需要根据本地用户的实际情况而定。本申请实施例基于相似度系数、时间差、接收顺位、亲密度和提交数量等多维参数共同确定对话信息的权重系数,可以使得最终确定的权重系数更加精确。
在上述实施例的基础上,参见图3所示,步骤102中“根据处理结果确定与目标对话信息相对应的一个或多个回复信息”包括步骤1021-1024:
步骤1021:将目标对话信息的处理结果输入至预设的信息匹配模型中,将信息匹配模型的输出结果作为与目标对话信息对应的临时回复信息;信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型。
本申请实施例中,预设神经网络模型,预先将多个经自然语言理解处理的目标对话信息作为输入、将用户回复该目标对话信息的回复信息作为输出训练神经网络模型,并得到神经网络模型的具体参数,将训练后的神经网络模型作为信息匹配模型。在获得本地用户需要回复的目标对话信息的处理结果之后,将该处理结果作为信息匹配模型的输入即可得到相应的用于回复该处理结果的信息,即临时回复信息。
步骤1022:判断临时回复信息中是否存在用于调用第三方数据库的调用标识,在临时回复信息中不存在调用标识时时,将临时回复信息作为与对话信息相对应的回复信息。
步骤1023:在临时回复信息中存在调用标识时时,查询与调用标识相应的第三方数据库,并根据查询结果补充临时回复信息,将补充后的临时回复信息作为与目标对话信息相对应的回复信息。
本申请实施例中,在确定回复信息时还可借助第三方数据库,该第三方数据库为需要实时更新的数据库,具体可以包括气象数据库、路况数据库、航班数据库、本地数据库等;本地数据库中存储有与本地用户相关的数据,如本地用户的行程表、备忘录等。该调用标识用于表示是否需要调用第三方数据库,当不需要调用第三方数据库时,该调用标识可以为null;当需要调用第三方数据库时,该调用标识与所需的第三方数据库相一致;一个回复信息可别包含多个调用标识。具体的,例如,其他用户提交的对话信息为“你那天气如何?”,此时除了依靠神经网络模型确定需要回复的信息外,还需要确定调用标识是否存在以及存在时是哪一调用标识,若调用标识存在且与气象数据库相对应,则此时借助气象数据库来确定当前的天气信息;或者先借助终端的定位信息确定本地用户的当前位置,进而借助气象数据库来确定当前位置当前时间的天气信息。当根据神经网络模型可以直接确定回复信息时,则可不需要调用第三方数据库。本申请实施例通过神经网络模型可以快速确定回复信息,同时在适当的情况下调用第三方数据库,能够提供更加完善精确的回复信息。
在上述实施例的基础上,在步骤103“将目标回复信息添加至目标对话界面的文本输入栏中”之后,该方法还包括后续处理的过程,该过程具体包括步骤B1-B2:
步骤B1:当接收到本地用户输入的修改指令时,根据修改指令修改回复信息,将修改后的回复信息作为有效回复信息并发送。
步骤B2:将目标对话信息作为输入、有效回复信息作为输出,训练并更新预设的信息匹配模型。
本申请实施例中,信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型,在将目标回复信息添加至文本输入栏中之后,本地用户可以直接发送该目标回复信息,也可以编辑该目标回复信息;具体的,本地用户可以输入修改指令来编辑该目标回复信息。当存在用户编辑目标回复信息的情况时,说明基于信息匹配模型确定的目标回复信息并不是完全适合该本地用户的,此时可以将本地用户修改后的回复信息(即有效回复信息)作为样本,继续训练神经网络模型,调整神经网络模型的参数,使得训练后的神经网络模型(即信息匹配模型)更加符合本地用户的需求。
本申请实施例提供的一种基于上下文的输入方法,在本地用户未执行任何输入操作之前即可自动获取非本地用户发送的目标对话信息,进而确定相应的回复消息并将其中的目标回复消息添加至输入栏,方便本地用户直接发送该目标回复消息或编辑目标回复消息。该方法在本地用户未执行任何输入操作之前即可自动生成目标回复消息,简化了用户操作,节省了操作时间,可以实现快速回复,提高了用户操作效率。基于对话信息的权重系数来 选定对本地用户比较重要的对话信息,从而方便后续优先对该重要的对话信息进行自动回复。基于多维参数共同确定对话信息的权重系数,可以使得最终确定的权重系数更加精确。通过神经网络模型可以快速确定回复信息,同时在适当的情况下调用第三方数据库,能够提供更加完善精确的回复信息;且基于本地用户修改后的有效回复信息训练神将网络模型,可以使得神经网络模型更加符合本地用户的需求。
以上详细介绍了基于上下文的输入的方法流程,该方法也可以通过相应的装置实现,下面详细介绍该装置的结构和功能。
本申请实施例提供一种基于上下文的输入装置,参见图4所示,包括:
获取模块41,用于当接收到本地用户在目标对话界面触发的信息输入指令时,获取目标对话界面中非本地用户发送的目标对话信息;
处理模块42,用于对目标对话信息进行自然语言理解处理,根据处理结果确定与目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
输入模块43,用于当接收到本地用户针对目标回复信息的选择指令时,将目标回复信息添加至目标对话界面的文本输入栏中。
在一种可能的实现方式中,参见图5所示,获取模块41包括:
时间确定单元411,用于将本地用户最新提交信息的时间点作为起始时间点;
获取单元412,用于确定在起始时间点与当前时间点之间所接收的非本地用户发送的所有对话信息,并确定每条对话信息对应的信息参数,信息参数包括信息内容、获取时间和其他用户的用户标识;
处理单元413,用于根据信息参数确定相应对话信息的权重系数,将权重系数最大或权重系数大于预设阈值的对话信息作为目标对话信息,目标对话信息对应的时间点为目标时间点。
在一种可能的实现方式中,处理单元413包括:
预处理子单元,用于将在起始时间点与当前时间点之间所获取的一个对话信息作为第一对话信息;
第一确定子单元,用于确定第一对话信息的信息内容与第二对话信息的信息内容之间的相似度,根据所有的相似度确定第一对话信息的相似度系数;第二对话信息为除第一对话信息以外的第二对话信息;
第二确定子单元,用于确定第一对话信息的获取时间与当前时间点之间的时间差,并根据第一对话信息的获取时间确定第一对话信息的接收顺位;
第三确定子单元,用于根据第一对话信息的用户标识确定相对应的非本地用户与本地用户之间的亲密度,并确定第一对话信息的用户标识在起始时间点与当前时间点之间提交对话信息的提交数量;
处理子单元,用于根据相似度系数、时间差、接收顺位、亲密度和提交数量确定第一对话信息的权重系数;权重系数与相似度系数、亲密度和提交数量呈正相关关系,权重系数与时间差和接收顺位的倒数呈正相关关系或负相关关系。
在一种可能的实现方式中,参见图6所示,处理模块42包括:
预设单元421,用于将所述目标对话信息的处理结果输入至预设的信息匹配模型中;
信息确定单元422,用于将所述信息匹配模型的输出结果作为与所述目标对话信息对应的临时回复信息;所述信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型;
处理单元423,用于判断临时回复信息中是否存在用于调用第三方数据库的调用标识,第三方数据库包括气象数据库、路况数据库、航班数据库、本地数据库中的一项或多项;在临时回复信息中不存在调用标识时,将临时回复信息作为与目标对话信息相对应的回复信息;在临时回复信息中存在调用标识时,查询与调用标识相应的第三方数据库,并根据查询结果补充临时回复信息,将补充后的临时回复信息作为与目标对话信息相对应的回复信息。
在一种可能的实现方式中,参见图7所示,该装置还包括修改模块44和训练模块45;
在输入模块43将目标回复信息添加至文本输入栏中之后,修改模块44用于当接收到本地用户输入的修改指令时,根据修改指令修改目标回复信息,将修改后的目标回复信息作为有效回复信息并发送;
训练模块45用于将目标对话信息作为输入、有效回复信息作为输出,训练并更新预设的信息匹配模型,信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型。
在一种可能的实现方式中,获取模块41具体用于:将目标对话界面内最后一条非本地用户发送的对话信息作为目标对话信息。
本申请实施例还提供了一种非易失性计算机可读存储介质,非易失性计算机可读存储介质存储有计算机可读指令,其包含用于执行上述基于上下文的输入方法的程序,该计算机可读指令可执行上述任意方法实施例中的方法。
其中,所述非易失性计算机可读存储介质可以是计算机能够存取的任何可用介质或数据存储设备,包括但不限于磁性存储器(例如软盘、硬盘、磁带、磁光盘(MO)等)、光学存储 器(例如CD、DVD、BD、HVD等)、以及半导体存储器(例如ROM、EPROM、EEPROM、非易失性存储器(NAND FLASH)、固态硬盘(SSD))等。
图8示出了本申请的另一个实施例的一种计算机设备的结构框图。所述计算机设备1100可以是具备计算能力的主机服务器、个人计算机PC、或者可携带的便携式计算机或计算机设备等。本申请具体实施例并不对计算机设备的具体实现做限定。
该计算机设备1100包括至少一个处理器(processor)1110、通信接口(Communications Interface)1120、存储器(memory array)1130和总线1140。其中,处理器1110、通信接口1120、以及存储器1130通过总线1140完成相互间的通信。
通信接口1120用于与网元通信,其中网元包括例如虚拟机管理中心、共享存储等。
处理器1110用于执行计算机可读指令。处理器1110可能是一个中央处理器CPU,或者是专用集成电路ASIC(Application Specific IntegratedCircuit),或者是被配置成实施本申请实施例的一个或多个集成电路。
存储器1130用于可执行的指令。存储器1130可能包含高速RAM存储器,也可能还包括非易失性存储器(non-volatile memory),例如至少一个磁盘存储器。存储器1130也可以是存储器阵列。存储器1130还可能被分块,并且所述块可按一定的规则组合成虚拟卷。存储器1130存储的计算机可读指令可被处理器1110执行,以使处理器1110能够执行上述任意方法实施例中的方法。
显然,本领域的技术人员可以对本申请进行各种改动和变型而不脱离本申请的精神和范围。这样,倘若本申请的这些修改和变型属于本申请权利要求及其等同技术的范围之内,则本申请也意图包含这些改动和变型在内。

Claims (20)

  1. 一种基于上下文的输入方法,其特征在于,包括:
    当接收到本地用户在目标对话界面触发的信息输入指令时,获取所述目标对话界面中非本地用户发送的目标对话信息;
    对所述目标对话信息进行自然语言理解处理,根据处理结果确定与所述目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
    当接收到所述本地用户针对目标回复信息的选择指令时,将所述目标回复信息添加至所述目标对话界面的文本输入栏中。
  2. 根据权利要求1所述的输入方法,其特征在于,所述获取所述目标对话界面中非本地用户发送的目标对话信息包括:
    将本地用户最新提交信息的时间点作为起始时间点;
    确定在所述起始时间点与所述当前时间点之间所接收的非本地用户发送的所有对话信息,并确定每条对话信息对应的信息参数,所述信息参数包括信息内容、获取时间和其他用户的用户标识;
    根据所述信息参数确定相应对话信息的权重系数,将权重系数最大或权重系数大于预设阈值的对话信息作为目标对话信息。
  3. 根据权利要求2所述的输入方法,其特征在于,所述根据所述信息参数确定相应对话信息的权重系数包括:
    将在所述起始时间点与所述当前时间点之间所获取的一个对话信息作为第一对话信息;
    确定所述第一对话信息的信息内容与第二对话信息的信息内容之间的相似度,根据所有的相似度确定所述第一对话信息的相似度系数;所述第二对话信息为除所述第一对话信息以外的第二对话信息;
    确定所述第一对话信息的获取时间与所述当前时间点之间的时间差,并根据所述第一对话信息的获取时间确定所述第一对话信息的接收顺位;
    根据所述第一对话信息的用户标识确定相对应的非本地用户与所述本地用户之间的亲密度,并确定所述第一对话信息的用户标识在所述起始时间点与所述当前时间点之间提交对话信息的提交数量;
    根据所述相似度系数、时间差、接收顺位、亲密度和提交数量确定所述第一对话信息 的权重系数;所述权重系数与所述相似度系数、所述亲密度和所述提交数量呈正相关关系,所述权重系数与所述时间差和所述接收顺位的倒数呈正相关关系或负相关关系。
  4. 根据权利要求1所述的输入方法,其特征在于,所述根据处理结果确定与所述目标对话信息相对应的一个或多个回复信息包括:
    将所述目标对话信息的处理结果输入至预设的信息匹配模型中,将所述信息匹配模型的输出结果作为与所述目标对话信息对应的临时回复信息;所述信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型;
    判断所述临时回复信息中是否存在用于调用第三方数据库的调用标识,所述第三方数据库包括气象数据库、路况数据库、航班数据库、本地数据库中的一项或多项;
    在所述临时回复信息中不存在调用标识时,将所述临时回复信息作为与所述目标对话信息相对应的回复信息;
    在所述临时回复信息中存在调用标识时,查询与所述调用标识相应的第三方数据库,并根据查询结果补充所述临时回复信息,将补充后的临时回复信息作为与所述目标对话信息相对应的回复信息。
  5. 根据权利要求1所述的输入方法,其特征在于,在所述将所述目标回复信息添加至所述目标对话界面的文本输入栏中之后,还包括:
    当接收到本地用户输入的修改指令时,根据所述修改指令修改所述目标回复信息,将修改后的目标回复信息作为有效回复信息并发送。
  6. 根据权利要求5所述的输入方法,其特征在于,在所述将修改后的目标回复信息作为有效回复信息之后,还包括:
    将所述目标对话信息作为输入、所述有效回复信息作为输出,训练并更新预设的信息匹配模型,所述信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型。
  7. 根据权利要求1所述的输入方法,其特征在于,所述获取所述目标对话界面中非本地用户发送的目标对话信息包括:
    将所述目标对话界面内最后一条非本地用户发送的对话信息作为目标对话信息。
  8. 一种基于上下文的输入装置,其特征在于,包括:
    获取模块,用于当接收到本地用户在目标对话界面触发的信息输入指令时,获取所述目标对话界面中非本地用户发送的目标对话信息;
    处理模块,用于对所述目标对话信息进行自然语言理解处理,根据处理结果确定与所 述目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
    输入模块,用于当接收到所述本地用户针对目标回复信息的选择指令时,将所述目标回复信息添加至所述目标对话界面的文本输入栏中。
  9. 根据权利要求8所述的装置,其特征在于,所述获取模块包括:
    时间确定单元,用于将本地用户最新提交信息的时间点作为起始时间点;
    获取单元,用于确定在所述起始时间点与所述当前时间点之间所接收的非本地用户发送的所有对话信息,并确定每条对话信息对应的信息参数,所述信息参数包括信息内容、获取时间和其他用户的用户标识;
    处理单元,用于根据所述信息参数确定相应对话信息的权重系数,将权重系数最大或权重系数大于预设阈值的对话信息作为目标对话信息。
  10. 根据权利要求9所述的装置,其特征在于,所述处理单元包括:
    预处理子单元,用于将在所述起始时间点与所述当前时间点之间所获取的一个对话信息作为第一对话信息;
    第一确定子单元,用于确定所述第一对话信息的信息内容与第二对话信息的信息内容之间的相似度,根据所有的相似度确定所述第一对话信息的相似度系数;所述第二对话信息为除所述第一对话信息以外的第二对话信息;
    第二确定子单元,用于确定所述第一对话信息的获取时间与所述当前时间点之间的时间差,并根据所述第一对话信息的获取时间确定所述第一对话信息的接收顺位;
    第三确定子单元,用于根据所述第一对话信息的用户标识确定相对应的非本地用户与所述本地用户之间的亲密度,并确定所述第一对话信息的用户标识在所述起始时间点与所述当前时间点之间提交对话信息的提交数量;
    处理子单元,用于根据所述相似度系数、时间差、接收顺位、亲密度和提交数量确定所述第一对话信息的权重系数;所述权重系数与所述相似度系数、所述亲密度和所述提交数量呈正相关关系,所述权重系数与所述时间差和所述接收顺位的倒数呈正相关关系或负相关关系。
  11. 根据权利要求8所述的装置,其特征在于,所述处理模块包括:
    预设单元,将所述目标对话信息的处理结果输入至预设的信息匹配模型中;
    信息确定单元,用于将所述信息匹配模型的输出结果作为与所述目标对话信息对应的临时回复信息;所述信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型;
    处理单元,用于判断所述临时回复信息中是否存在用于调用第三方数据库的调用标识,所述第三方数据库包括气象数据库、路况数据库、航班数据库、本地数据库中的一项或多项;在所述临时回复信息中不存在调用标识时,将所述临时回复信息作为与所述目标对话信息相对应的回复信息;在所述临时回复信息中存在调用标识时,查询与所述调用标识相应的第三方数据库,并根据查询结果补充所述临时回复信息,将补充后的临时回复信息作为与所述目标对话信息相对应的回复信息。
  12. 根据权利要求8所述的装置,其特征在于,还包括修改模块;
    在所述输入模块将目标回复信息添加至文本输入栏中之后,所述修改模块用于当接收到本地用户输入的修改指令时,根据所述修改指令修改所述目标回复信息,将修改后的目标回复信息作为有效回复信息并发送。
  13. 根据权利要求12所述的装置,其特征在于,还包括训练模块;
    所述训练模块用于将所述目标对话信息作为输入、所述有效回复信息作为输出,训练并更新预设的信息匹配模型,所述信息匹配模型为多个经自然语言理解处理的目标对话信息和对应的回复信息作为样本输入神经网络模型后训练得到的模型。
  14. 根据权利要求8所述的装置,其特征在于,所述获取模块用于:将所述目标对话界面内最后一条非本地用户发送的对话信息作为目标对话信息。
  15. 一种非易失性计算机可读存储介质,所述非易失性计算机可读存储介质中存储有至少一计算机可读指令,所述计算机可读指令被处理器执行时实现基于上下文的输入方法,所述方法包括:
    当接收到本地用户在目标对话界面触发的信息输入指令时,获取所述目标对话界面中非本地用户发送的目标对话信息;
    对所述目标对话信息进行自然语言理解处理,根据处理结果确定与所述目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
    当接收到所述本地用户针对目标回复信息的选择指令时,将所述目标回复信息添加至所述目标对话界面的文本输入栏中。
  16. 根据权利要求15所述的非易失性计算机可读存储介质,其特征在于,所述计算机可读指令被处理器执行时实现所述获取所述目标对话界面中非本地用户发送的目标对话信息,包括:
    将本地用户最新提交信息的时间点作为起始时间点;
    确定在所述起始时间点与所述当前时间点之间所接收的非本地用户发送的所有对话信 息,并确定每条对话信息对应的信息参数,所述信息参数包括信息内容、获取时间和其他用户的用户标识;
    根据所述信息参数确定相应对话信息的权重系数,将权重系数最大或权重系数大于预设阈值的对话信息作为目标对话信息。
  17. 根据权利要求16所述的非易失性计算机可读存储介质,其特征在于,所述计算机可读指令被处理器执行时实现所述根据所述信息参数确定相应对话信息的权重系数,包括:
    将在所述起始时间点与所述当前时间点之间所获取的一个对话信息作为第一对话信息;
    确定所述第一对话信息的信息内容与第二对话信息的信息内容之间的相似度,根据所有的相似度确定所述第一对话信息的相似度系数;所述第二对话信息为除所述第一对话信息以外的第二对话信息;
    确定所述第一对话信息的获取时间与所述当前时间点之间的时间差,并根据所述第一对话信息的获取时间确定所述第一对话信息的接收顺位;
    根据所述第一对话信息的用户标识确定相对应的非本地用户与所述本地用户之间的亲密度,并确定所述第一对话信息的用户标识在所述起始时间点与所述当前时间点之间提交对话信息的提交数量;
    根据所述相似度系数、时间差、接收顺位、亲密度和提交数量确定所述第一对话信息的权重系数;所述权重系数与所述相似度系数、所述亲密度和所述提交数量呈正相关关系,所述权重系数与所述时间差和所述接收顺位的倒数呈正相关关系或负相关关系。
  18. 一种计算机设备,包括:处理器、存储器、通信接口和通信总线,所述处理器、所述存储器和所述通信接口通过所述通信总线完成相互间的通信;
    所述存储器用于存放至少一计算机可读指令,所述处理器执行所述计算机可读指令时实现基于上下文的输入方法,包括:
    当接收到本地用户在目标对话界面触发的信息输入指令时,获取所述目标对话界面中非本地用户发送的目标对话信息;
    对所述目标对话信息进行自然语言理解处理,根据处理结果确定与所述目标对话信息相对应的一个或多个回复信息并展示,所述处理结果包括与所述目标对话信息对应的特征词的有序序列;
    当接收到所述本地用户针对目标回复信息的选择指令时,将所述目标回复信息添加至所述目标对话界面的文本输入栏中。
  19. 根据权利要求18所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时实现所述获取所述目标对话界面中非本地用户发送的目标对话信息,包括:
    将本地用户最新提交信息的时间点作为起始时间点;
    确定在所述起始时间点与所述当前时间点之间所接收的非本地用户发送的所有对话信息,并确定每条对话信息对应的信息参数,所述信息参数包括信息内容、获取时间和其他用户的用户标识;
    根据所述信息参数确定相应对话信息的权重系数,将权重系数最大或权重系数大于预设阈值的对话信息作为目标对话信息。
  20. 根据权利要求19所述的计算机设备,其特征在于,所述处理器执行所述计算机可读指令时实现所述根据所述信息参数确定相应对话信息的权重系数,包括:
    将在所述起始时间点与所述当前时间点之间所获取的一个对话信息作为第一对话信息;
    确定所述第一对话信息的信息内容与第二对话信息的信息内容之间的相似度,根据所有的相似度确定所述第一对话信息的相似度系数;所述第二对话信息为除所述第一对话信息以外的第二对话信息;
    确定所述第一对话信息的获取时间与所述当前时间点之间的时间差,并根据所述第一对话信息的获取时间确定所述第一对话信息的接收顺位;
    根据所述第一对话信息的用户标识确定相对应的非本地用户与所述本地用户之间的亲密度,并确定所述第一对话信息的用户标识在所述起始时间点与所述当前时间点之间提交对话信息的提交数量;
    根据所述相似度系数、时间差、接收顺位、亲密度和提交数量确定所述第一对话信息的权重系数;所述权重系数与所述相似度系数、所述亲密度和所述提交数量呈正相关关系,所述权重系数与所述时间差和所述接收顺位的倒数呈正相关关系或负相关关系。
PCT/CN2018/111681 2018-08-01 2018-10-24 基于上下文的输入方法、装置、存储介质及计算机设备 WO2020024455A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810864174.0 2018-08-01
CN201810864174.0A CN109032381B (zh) 2018-08-01 2018-08-01 一种基于上下文的输入方法、装置、存储介质及终端

Publications (1)

Publication Number Publication Date
WO2020024455A1 true WO2020024455A1 (zh) 2020-02-06

Family

ID=64647556

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/111681 WO2020024455A1 (zh) 2018-08-01 2018-10-24 基于上下文的输入方法、装置、存储介质及计算机设备

Country Status (2)

Country Link
CN (1) CN109032381B (zh)
WO (1) WO2020024455A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111522933A (zh) * 2020-04-23 2020-08-11 深圳追一科技有限公司 一种对话流程控制方法、装置、设备及介质
CN111581623A (zh) * 2020-05-09 2020-08-25 深圳物控智联科技有限公司 智能数据交互方法、装置、电子设备及存储介质
CN112035613A (zh) * 2020-08-28 2020-12-04 光大科技有限公司 基于多轮对话的信息推送方法及装置、存储介质、电子装置
CN113242172A (zh) * 2021-04-26 2021-08-10 福建天泉教育科技有限公司 一种消息的应答方法及系统
CN113300941A (zh) * 2021-05-20 2021-08-24 维沃移动通信(杭州)有限公司 显示方法、显示装置、相关设备及可读存储介质
CN113887210A (zh) * 2021-10-19 2022-01-04 京东科技信息技术有限公司 时间信息的处理方法、装置、电子设备及存储介质
CN117131182A (zh) * 2023-10-26 2023-11-28 江西拓世智能科技股份有限公司 一种基于ai的对话回复生成方法及系统

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113794623A (zh) * 2021-08-31 2021-12-14 北京明略软件系统有限公司 一种应答消息的生成方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080046394A1 (en) * 2006-08-14 2008-02-21 Microsoft Corporation Knowledge extraction from online discussion forums
CN104750705A (zh) * 2013-12-27 2015-07-01 华为技术有限公司 信息回复方法及装置
CN106997375A (zh) * 2017-02-28 2017-08-01 浙江大学 基于深度学习的客服回复推荐方法
CN108021572A (zh) * 2016-11-02 2018-05-11 腾讯科技(深圳)有限公司 回复信息推荐方法和装置
CN108121799A (zh) * 2017-12-21 2018-06-05 广东欧珀移动通信有限公司 回复语句的推荐方法、装置、存储介质及移动终端

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10192551B2 (en) * 2016-08-30 2019-01-29 Google Llc Using textual input and user state information to generate reply content to present in response to the textual input
CN107168546B (zh) * 2017-03-27 2021-03-09 上海奔影网络科技有限公司 输入提示方法及装置
CN107612814A (zh) * 2017-09-08 2018-01-19 北京百度网讯科技有限公司 用于生成候选回复信息的方法和装置
CN107566256B (zh) * 2017-09-13 2019-04-12 Oppo广东移动通信有限公司 信息回复方法、装置、终端和计算机可读存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080046394A1 (en) * 2006-08-14 2008-02-21 Microsoft Corporation Knowledge extraction from online discussion forums
CN104750705A (zh) * 2013-12-27 2015-07-01 华为技术有限公司 信息回复方法及装置
CN108021572A (zh) * 2016-11-02 2018-05-11 腾讯科技(深圳)有限公司 回复信息推荐方法和装置
CN106997375A (zh) * 2017-02-28 2017-08-01 浙江大学 基于深度学习的客服回复推荐方法
CN108121799A (zh) * 2017-12-21 2018-06-05 广东欧珀移动通信有限公司 回复语句的推荐方法、装置、存储介质及移动终端

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111522933A (zh) * 2020-04-23 2020-08-11 深圳追一科技有限公司 一种对话流程控制方法、装置、设备及介质
CN111581623A (zh) * 2020-05-09 2020-08-25 深圳物控智联科技有限公司 智能数据交互方法、装置、电子设备及存储介质
CN111581623B (zh) * 2020-05-09 2023-12-19 深圳物控智联科技有限公司 智能数据交互方法、装置、电子设备及存储介质
CN112035613A (zh) * 2020-08-28 2020-12-04 光大科技有限公司 基于多轮对话的信息推送方法及装置、存储介质、电子装置
CN112035613B (zh) * 2020-08-28 2023-09-12 光大科技有限公司 基于多轮对话的信息推送方法及装置、存储介质、电子装置
CN113242172A (zh) * 2021-04-26 2021-08-10 福建天泉教育科技有限公司 一种消息的应答方法及系统
CN113300941A (zh) * 2021-05-20 2021-08-24 维沃移动通信(杭州)有限公司 显示方法、显示装置、相关设备及可读存储介质
CN113300941B (zh) * 2021-05-20 2023-04-18 维沃移动通信(杭州)有限公司 显示方法、显示装置、相关设备及可读存储介质
CN113887210A (zh) * 2021-10-19 2022-01-04 京东科技信息技术有限公司 时间信息的处理方法、装置、电子设备及存储介质
CN117131182A (zh) * 2023-10-26 2023-11-28 江西拓世智能科技股份有限公司 一种基于ai的对话回复生成方法及系统

Also Published As

Publication number Publication date
CN109032381B (zh) 2022-05-17
CN109032381A (zh) 2018-12-18

Similar Documents

Publication Publication Date Title
WO2020024455A1 (zh) 基于上下文的输入方法、装置、存储介质及计算机设备
US11216510B2 (en) Processing an incomplete message with a neural network to generate suggested messages
US10230668B2 (en) Information replying method and apparatus
US10628474B2 (en) Probabalistic generation of diverse summaries
CN112270925B (zh) 用于创建可定制对话系统引擎的平台
US9582757B1 (en) Scalable curation system
CN110888990B (zh) 文本推荐方法、装置、设备及介质
WO2018024166A1 (zh) 确定候选输入的方法、输入提示方法和电子设备
CN111881316B (zh) 搜索方法、装置及服务器和计算机可读存储介质
JP2019503526A (ja) ダイアログ・システムにおけるパラメータ収集および自動ダイアログ生成
WO2022134421A1 (zh) 基于多知识图谱的智能答复方法、装置、计算机设备及存储介质
CN117172288A (zh) 合作地训练和/或使用单独的输入神经网络模型和响应神经网络模型以用于确定针对电子通信的响应
CN107844470B (zh) 一种语音数据处理方法及其设备
CN110147494B (zh) 信息搜索方法、装置,存储介质及电子设备
US11379527B2 (en) Sibling search queries
CN110287364B (zh) 语音搜索方法、系统、设备及计算机可读存储介质
CN107862004A (zh) 智能排序方法及装置、存储介质、电子设备
WO2020233381A1 (zh) 基于语音识别的服务请求方法、装置及计算机设备
CN116501960B (zh) 内容检索方法、装置、设备及介质
KR101695348B1 (ko) 메신저 기반 서비스 제공 장치 및 이를 이용한 방법
CN111159334A (zh) 用于房源跟进信息处理的方法及系统
CN108306813B (zh) 会话消息的处理方法、服务器及客户端
CN109918661B (zh) 同义词获取方法及装置
CN112069803A (zh) 文本备份方法、装置、设备及计算机可读存储介质
CN111159526B (zh) 查询语句处理方法、装置、设备及存储介质

Legal Events

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

Ref document number: 18928710

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18928710

Country of ref document: EP

Kind code of ref document: A1