Automatic message reply method, device, system, electronic device and readable medium

Abstract

The present invention discloses an automatic message reply method, device, system, electronic device and readable medium. In the method, in response to a target event triggered by a local chat tool, in the target event, the chat tool is associated with the automatic reply configuration in the input method application in real time; in response to the first message triggered by the chat tool, the conversation processing model generated by training is used to identify the corpus information of the text or picture in the first message, the chat intent in the corpus information is parsed and obtained, and the reply content adapted to the chat intent is generated and output; in response to the reply content generated by the input method application, the chat tool uses the reply content as a second message to perform automatic reply. The present invention achieves the purpose of remote human-computer interaction based on the chat tool by loading the automatic message reply in the input method application, and expands the field of human-computer interaction for non-commercial purposes.

Description

Technical Field

[0001] The present invention relates to the field of input method application technology, and in particular to an automatic message reply method, device, system, electronic device and readable medium. Background Art

[0002] With the rapid development of Internet technology, the market direction of products or services has used intelligent customer service assistants to identify keywords in consultation information and select more accurate responses from many candidate answers. Therefore, conversational robots that provide personalized services to users have become the research and development direction and focus of major companies and academic research institutions.

[0003] However, the inventor of this application has found that the purpose of existing conversational robots is mostly to sell products or services better, and they cannot help ordinary people communicate with each other. If the conversational robots on the market are directly applied to communication between ordinary people, it is necessary to build various answers for various questions and accurately design the selection logic, which requires huge manpower, and the design of dependencies between contexts will be very complicated, and there will be conflicts between rules. If the reply in the conversation scene is based on the intention of the corpus, the reply information will be unsatisfactory and cannot achieve the purpose of communication. Most of the existing conversational robots are embedded in the design of chat tools, and no technical solutions related to intelligent chat have been found in the field of input methods. Summary of the invention

[0004] In order to solve the technical problems in the background technology, the embodiments of the present application provide an automatic message reply method, device, system, electronic device and readable medium, and achieve the purpose of remote human-computer interaction based on a chat tool by loading the automatic message reply in an input method application, and expand the field of human-computer interaction for non-commercial purposes.

[0005] In a first aspect, an embodiment of the present application provides a method for automatically replying to a message, the method comprising:

[0006] In response to a target event triggered by a local chat tool, in the target event, the chat tool is associated with an automatic reply configuration in an input method application in real time;

[0007] In response to a first message received by the chat tool, after identifying the text or image corpus information in the first message using the trained dialogue processing model, the chat intent in the corpus information is parsed and obtained, and a reply content adapted to the chat intent is generated and output;

[0008] In response to the reply content generated by the input method application, the chat tool uses the reply content as a second message to perform an automatic reply.

[0009] Furthermore, in response to the input method application generating the second message based on the first message in the local chat tool, the second message is automatically input into the local chat tool for transmission to the remote chat tool, so that the remote user can perform remote human-computer interaction with the input method application based on the chat tools at both ends.

[0010] Furthermore, in the input method application, in addition to the automatic reply configuration, a user input configuration is also provided;

[0011] When the target event triggered in the local chat tool is the user input configuration associated with the input method application, the user's information input is received as a second message to execute a reply.

[0012] Furthermore, in response to a target event triggered by the local chat tool, the information input configuration triggered by the local user is replaced by the automatic reply configuration in the input method application.

[0013] Furthermore, the dialogue processing model is formed by training a large amount of dialogue data using a Transformer network architecture. Furthermore, when the dialogue processing model is used to identify the corpus information of the text or picture in the first message, the text or picture in the first message is cleaned and segmented using natural language processing technology and text recognition technology to obtain valid corpus elements including single words and phrases, and other element information other than the valid corpus elements in the first message is removed to meet the parsing requirements.

[0014] In a second aspect, an embodiment of the present application provides an automatic message reply device, comprising:

[0015] An association module, configured to respond to a target event triggered by a local chat tool, wherein the chat tool is associated with an automatic reply configuration in an input method application in real time in the target event;

[0016] A dialogue processing module is configured to respond to a first message triggered by the chat tool, identify text or image corpus information in the first message using a trained dialogue processing model, parse and obtain chat intent in the corpus information, and generate and output reply content adapted to the chat intent;

[0017] The information reply module is configured to respond to the reply content generated by the input method application and perform automatic reply via the chat tool using the reply content as the second message.

[0018] In a third aspect, an embodiment of the present application provides an automatic message reply system, comprising: a remote user terminal, a local user terminal and a cloud, wherein the cloud is respectively connected to the remote user terminal and the local user terminal in communication; the remote user terminal and the local user terminal are both configured with a chat tool and an input method application;

[0019] The cloud receives a first message in response to the remote user calling a user input configuration in an input method application in the chat tool of the remote user terminal, and feeds the first message back to the local user terminal, and receives a second message replied in the local user terminal and feeds the second message back to the remote user terminal;

[0020] Among them, the local user end responds to a target event triggered by a local chat tool, in which the chat tool associates the automatic reply configuration in the input method application in real time; in response to the first message triggered and received by the chat tool, the conversation processing model generated by training is used to identify the corpus information of the text or picture in the first message, the chat intent in the corpus information is parsed and obtained, and the reply content adapted to the chat intent is generated and output; in response to the reply content generated by the input method application, the reply content is used as the second message to perform an automatic reply through the chat tool.

[0021] In a fourth aspect, an embodiment of the present application provides an electronic device, including:

[0022] one or more processors;

[0023] A memory for storing one or more programs;

[0024] When the one or more programs are executed by the one or more processors, the one or more processors implement any method as described in the first aspect.

[0025] In a fifth aspect, an embodiment of the present application provides a computer-readable medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, it implements any method described in the first aspect.

[0026] The multiple technical solutions provided in the embodiments of the present application have at least the following technical effects:

[0027] 1. Since the dialogue processing model is loaded into the input method program and the purpose of remote human-computer interaction is achieved through a chat tool, compared with the existing input method technology that solves the problem of "electronicizing" language expression symbols, it is convenient for users to record or communicate on electronic devices without an input carrier based on the user's selectable automatic reply function. In the modern society where time is fragmented, using the input method program to configure automatic replies can help people automatically reply to messages when local users are unable to reply to each other's messages in time, and can automatically reply with emoticons to promote communication.

[0028] 2. Due to the use of the Transformer model, text and emoticons can be automatically generated; that is, while obtaining the features of the text, the features of the image are also obtained, and then text and emoticons can be generated at the same time for automatic replies. And when training the model, the tedious data annotation work such as "labeling" emoticons is avoided, saving manpower and material resources.

[0029] 3. The chat content is based on different application scenarios and is not limited to small talk, task-based, and dialogue-based. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic diagram of the interaction of the existing chat tool in the first embodiment of the present application;

[0031] Figure 2 This is a flow chart of the automatic message reply method in Example 1 of the present application;

[0032] Figure 3 This is a module diagram of the automatic message reply device in Example 2 of the present application;

[0033] Figure 4 This is a schematic diagram of the automatic message reply system in Example 3 of the present application. DETAILED DESCRIPTION

[0034] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0035] Embodiment 1

[0036] The embodiment of the present application provides a method for automatically replying to messages. Before describing the method for automatically replying to messages in the present embodiment, the chat tool and the input method application are described separately. The chat tool can be understood as a conventional instant messaging tool, such as QQ, WeChat, Facebook Messenger, Facebook Messenger, etc. When using the chat tool to perform instant communication, the local user terminal and the remote user terminal are connected through the server (cloud) to which the chat tool belongs. The local user terminal and the remote user terminal are loaded with the same chat tool, and the input method application is not restricted. The present embodiment uses the local user terminal to express the automatic reply of messages.

[0037] Reference Figure 1 As shown, when using the chat tool to perform normal instant chat, the user of the local user end and the user of the remote user end each use the corresponding input method application to input information. The local user calls and loads the input method application through the chat tool on the local user end. After inputting the chat information, it is mapped to the chat tool. The corresponding remote user calls and loads the input method application through the chat tool on the remote user end. After the remote user inputs the chat information, it is also mapped to the chat tool. The local user and the remote user can understand each other's intentions through the corpus information of both parties in the chat tool. The embodiment improves this chat process by configuring a human-computer interaction design in the input method application.

[0038] Reference Figure 2 As shown, the automatic message reply method of this embodiment includes the following steps.

[0039] Step S100, in response to a target event triggered by a local chat tool, in the target event, the chat tool is associated with the automatic reply configuration in the input method application in real time.

[0040] In this embodiment, the input method application and the chat tool are improved, and the chat tool can choose to execute the automatic reply configuration in the real-time associated input method application. That is to say, the input method application can perform user input configuration, and can also perform automatic reply configuration. In this embodiment, in addition to the automatic reply configuration, a user input configuration is also provided in the input method application; when the target event triggered in the local chat tool is in response to the user input configuration of the associated input method application, the user's information input is received as the second message to execute the reply. When the chat tool is associated with the automatic reply configuration in the input method application in real time, the technical solution of the automatic information reply method in this embodiment is executed.

[0041] In the real-time automatic message reply method, in response to a target event triggered by the local chat tool, the automatic reply configuration in the input method application replaces the message input configuration triggered by the local user.

[0042] Step S200, in response to the first message triggered by the chat tool, after using the trained dialogue processing model to identify the corpus information of the text or picture in the first message, parse and obtain the chat intent in the corpus information, and generate and output reply content suitable for the chat intent.

[0043] In this embodiment, the dialogue processing model is formed by training a large amount of dialogue corpus using the Transformer network architecture. Of course, the dialogue processing model is certainly not limited to the Transformer network architecture. In this embodiment, in order to achieve the best effect of the model, the Transformer network architecture is used for model training. Before the dialogue processing model generates reply content according to the corpus information, a pre-built Transformer network is required to train and generate the required dialogue processing model. The Transformer network is different from the existing CNN convolutional neural network and RNN recurrent neural network. In order to train the model to achieve SOTA effect, the Transformer network architecture is directly used. On the one hand, it can solve the problem of slow training in the neural network, and on the other hand, it can perform fast parallelization of the self-attention mechanism, and can be increased to a very deep depth, giving full play to the characteristics of the deep learning model and improving the accuracy of model processing.

[0044] The Transformer network in this embodiment includes an input layer, through which valid corpus elements in the first message are obtained.

[0045] That is to say, when the embodiment uses the dialogue processing model to identify the corpus information of the text or picture in the first message, the natural language processing technology and the text recognition technology are used to clean and segment the text or picture in the first message to obtain valid corpus elements including single words and words, and other element information other than the valid corpus elements in the first message is eliminated to meet the parsing requirements. Further explanation: the corpus information in the first message is processed in the input layer to obtain the valid corpus elements in the first message. It can be seen that the purpose of performing cleaning and segmentation on the first message is to obtain the target format required for model training or recognition. For example, if the first message is "What are you doing?", then the cleaning process first removes the punctuation mark "?" and uses the natural language processing technology to separate it into "you" and "what are you doing", with spaces in between. If the first message is directly a picture, then after identifying the text information in the picture or extracting the text information associated with the picture, natural language processing and segmentation are performed.

[0046] To further explain, in this embodiment, since the dialogue processing model is set in the input method application, when obtaining the data sets required for model training, such as training sets and test sets, the application background of the input method application can be directly used based on natural language processing technology to perform network crawling of relevant data sets that are adapted to the language form of human logical thinking. This embodiment does not describe them one by one here, and any existing conventional data set collection technology can be used.

[0047] The Transformer network in this embodiment includes a coding layer, and the coding layer is configured with multiple encoders with the same structure connected end to end. Each encoder includes a self-attention sublayer and a feedforward neural network sublayer according to the element flow. In this embodiment, six encoders can be set, and a self-attention sublayer and a feedforward neural network sublayer are set in each encoder. The corpus element vectors in the vector list pass through the self-attention sublayer and the feedforward neural network sublayer in each encoder in parallel.

[0048] Among them, in the self-attention sub-layer, the matrix is ​​used to realize the self-attention calculation of each corpus element, which specifically includes the following steps:

[0049] Step S210, using the word embedding algorithm to convert the corpus elements in the corpus information into word vectors, a vector list is formed according to the input sequence, and each corpus element is embedded in a 512-dimensional vector. It can be seen that after each corpus element in the input sequence is word-embedded, no corpus element will pass through the self-attention sublayer and the feedforward neural network sublayer in the encoder. In the self-attention sublayer, each corpus element has its own path flow, and there is a dependency relationship between these path flows, while in the feedforward neural network sublayer, there is no dependency relationship, so in the feedforward neural network sublayer, each corpus element executes various paths in parallel.

[0050] In step S220, the self-attention sublayer in the encoder receives word vectors of multiple corpus elements at the same time. Each word vector is multiplied by three preset weight matrices to obtain the query vector, key vector, and value vector of each corpus element. The three new vectors are lower in dimension than the word vector, with a dimension of 64, while the word vector and the input / output vector of the encoder can keep most of the calculations of multi-head attention unchanged.

[0051] For example, the valid corpus information is {corpus element 1, corpus element 2}, the word vector X1 of corpus element 1, the word vector X2 of corpus element 2, the query vector weight matrix WQ, the key vector weight matrix WK, the value vector weight matrix WV, and the query vector Q1 of corpus element 1 is obtained by multiplying the word vector X1 with the query vector weight WQ. Similarly, the key vector K1 and value vector V1 of corpus element 1 are obtained. The query vector Q2, key vector K2 and value vector V2 are calculated in the corresponding corpus element 2.

[0052] After calculating the query vector Q1, key vector K1 and value vector V1 in corpus element 1, self-attention scoring is performed on corpus element 1. Based on the logic of the natural language processing algorithm, according to the input sequence, it can be seen that the attention value of each corpus element is different. In this embodiment, the score of corpus element 1 can be obtained by calculating the dot product of the key vector and query vector of corpus element 1. For example, the score of corpus element 1 is Q1*K1, and the score of corpus element 2 is Q2*K2.

[0053] To further illustrate, during training, it can be represented as a query matrix, a key matrix, and a value matrix. The word vector of corpus element 1 is input into the matrix X1 and multiplied by the trained weight matrix (WQ, WK, WV).

[0054] Step S230, divide the score of each corpus element by a predetermined parameter value, and then pass the result through softmax. Softmax means that the scores of each corpus element are normalized by the softmax function, and the scores of the corpus elements in the first message are all positive and the sum is 1. The predetermined parameter value can be the square root of the key vector dimension. In this embodiment, the square root of the key vector dimension 64 is 8, so as to achieve a more stable encoding gradient.

[0055] Among them, it can be seen that the softmax score determines the contribution of each corpus element to the current encoding position. On the one hand, the highest softmax score can be obtained, while also paying attention to the scores of the corpus elements related to it.

[0056] Step S240, multiply the word vector of each corpus element by the softmax score, so as to focus on semantically related corpus elements through the dependency relationship on the corpus element path flow.

[0057] Step S250, summing the weighted values, represented by parallel weighted summing of the word vectors of all corpus elements in the input sequence. The weight of each corpus element can be obtained by performing a dot product between the corresponding key vector and the query vector, and then performing a softmax calculation, so as to obtain the output of each corpus element at the input sequence position in the self-attention sublayer.

[0058] To further illustrate, the attention heads of all corpus elements are concatenated and multiplied by the weight matrix, which is jointly trained in the model. The result is a matrix that integrates the information of all attention heads, and the matrix result is passed to the feedforward neural network sublayer.

[0059] In addition to the encoding layer, the Transformer network is also provided with a decoding layer, which is configured with multiple decoders of the same structure connected end to end. Each decoder includes a self-attention sublayer, an encoding-decoding self-attention sublayer and a feedforward neural network sublayer according to the element flow. The encoder starts working by processing the input sequence. The output of the top encoder is then converted into an attention vector set containing a key vector K and a value vector V. These vectors are used by each decoder for its own "encoding-decoding attention sublayer". After completing the encoding phase, the decoding phase begins. Each step in the decoding phase will output an element of the output sequence, and it will be repeated until all corpus elements are decoded. The self-attention layer in the decoder behaves in a different mode from the encoder: in the decoder, the self-attention layer is only allowed to process those positions that are closer to the front in the output sequence. Before the softmax step, it will hide the positions behind. The encoding-decoding attention sublayer works basically like a multi-head self-attention layer, creating a query matrix through the next layer, and obtaining a key / value matrix from the output of the encoder. The decoding layer in this embodiment will finally output a real vector.

[0060] The Transformer network includes a linear transformation and a Softmax layer. Based on the real vector output in the decoding layer, the floating point numbers of the real vector are converted into corpus elements using a linear transformation. The linear transformation is based on a fully connected neural network, which can map the real vector in the decoding layer to a much larger vector called logits. The Softmax layer converts those scores into probabilities (all positive numbers with an upper limit of 1.0). The one with the highest probability is selected, and its corresponding reply corpus element is output.

[0061] It can be seen that the weights of the models are randomly generated. The probabilities generated by the untrained model are distributed in each corpus element and are assigned random values. The reply content is compared and the back-propagation algorithm is used to slightly adjust the weights of all models to generate reply content that is closer to the result.

[0062] Step S300, in response to the reply content generated by the input method application, automatically replying by using the chat tool using the reply content as a second message.

[0063] It can be seen that in this embodiment, in response to the input method application generating the second message based on the first message in the local chat tool, the second message is automatically input into the local chat tool for transmission to the remote chat tool, so that the remote user can perform remote human-computer interaction with the input method application based on the chat tools at both ends.

[0064] Embodiment 2

[0065] Reference Figure 3 As shown, an embodiment of the present application provides an automatic message reply device, which adopts the method in embodiment 1 and includes the following modules.

[0066] The association module 100 is configured to respond to a target event triggered by a local chat tool, in which the chat tool is associated with the automatic reply configuration in the input method application in real time.

[0067] The conversation processing module 200 is configured to respond to the first message triggered by the chat tool, use the trained conversation processing model to identify the corpus information of the text or picture in the first message, parse and obtain the chat intent in the corpus information, and generate and output reply content suitable for the chat intent.

[0068] The information reply module 300 is configured to respond to the reply content generated by the input method application and perform automatic reply via the chat tool using the reply content as a second message.

[0069] Embodiment 3

[0070] Reference Figure 4 As shown, an embodiment of the present application provides an automatic message reply system, which adopts the method in embodiment one and includes the following: a remote user terminal, a local user terminal and a cloud, and the cloud terminal is respectively communicated with the remote user terminal and the local user terminal.

[0071] Both the remote client and the local client are equipped with a chat tool and an input method application program.

[0072] The cloud obtains a first message in response to the remote user calling the user input configuration in the input method application in the chat tool of the remote user terminal, and feeds back the first message to the local user terminal, and receives a second message replied in the local user terminal and feeds back the second message to the remote user terminal.

[0073] Among them, the local user end responds to a target event triggered by a local chat tool, in which the chat tool associates the automatic reply configuration in the input method application in real time; in response to the first message triggered and received by the chat tool, after using the trained and generated dialogue processing model to identify the corpus information of the text or picture in the first message, the chat intent in the corpus information is parsed and obtained, and reply content suitable for the chat intent is generated and output; in response to the reply content generated by the input method application, the reply content is used as the second message to perform automatic reply through the chat tool.

[0074] Embodiment 4

[0075] In one embodiment, this embodiment provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement any method described in Example 1.

[0076] In another embodiment, this embodiment provides a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method described in any one of the embodiments is implemented.

[0077] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, devices, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0078] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0079] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0080] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0081] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0082] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for automatically replying information, characterized in that: The method comprises: responding to a target event triggered by a local chat tool, in which the chat tool is associated with an automatic reply configuration in an input method application in real time; responding to a first message received by the chat tool, identifying the text or image corpus information in the first message using a dialogue processing model generated through training, parsing and obtaining the chat intent in the corpus information, generating and outputting reply content adapted to the chat intent, wherein the dialogue processing model is formed by training a large amount of dialogue corpus using a Transformer network architecture, and the steps comprise: obtaining valid corpus elements in the first message through an input layer, converting the corpus elements into word vectors through a word embedding algorithm, forming a vector list according to an input sequence, and passing the word vectors in the vector list through a plurality of encoders with the same structure in sequence through an encoding layer, wherein the encoder comprises a self-attention sublayer, receiving a plurality of the word vectors and obtaining a query vector, a key vector, and a value vector of each of the corpus elements by multiplying them with three preset weight matrices; dividing the score of each of the corpus elements by a predetermined parameter value, normalize each score by using the softmax function, obtain the softmax scores of the corpus elements in the first message are all positive and the sum is 1, multiply the word vector of each corpus element by the softmax score, and perform weighted sum in parallel; obtain the output of each corpus element at the position of the input sequence in the self-attention sublayer; concatenate the attention heads of all the corpus elements and multiply them by the weight matrix, and transmit the matrix result to the feedforward neural network sublayer and decode and output the real vector through the decoding layer; based on the real vector output in the decoding layer, use a linear transformation to convert the floating point number of the real vector into the corpus element, wherein the linear transformation is based on a fully connected neural network, maps the real vector in the decoding layer to a logarithmic probability, passes through the Softmax layer and determines the corpus element with the highest probability, and outputs its corresponding reply corpus element; in response to the reply content generated by the input method application, perform automatic reply by using the chat tool with the reply content as the second message.

2. The automatic message reply method according to claim 1, characterized in that: In response to the input method application generating the second message based on the first message in the local chat tool, the second message is automatically input into the local chat tool for transmission to the remote chat tool, so that the remote user can perform remote human-computer interaction with the input method application based on the chat tools at both ends.

3. The automatic message reply method according to claim 1, characterized in that: In the input method application, in addition to the automatic reply configuration, a user input configuration is also provided; when responding to the target event triggered in the local chat tool as the user input configuration of the associated input method application, the user's information input is received as the second message to execute a reply.

4. The automatic message reply method according to claim 3, characterized in that: In response to a target event triggered by the local chat tool, the information input configuration triggered by the local user is replaced by the automatic reply configuration in the input method application.

5. The automatic message reply method according to claim 1, characterized in that: When the dialogue processing model is used to identify the corpus information of the text or pictures in the first message, the natural language processing technology and text recognition technology are used to clean and segment the text or pictures in the first message to obtain the valid corpus elements including single words and phrases, and other element information other than the valid corpus elements in the first message is eliminated to meet the parsing requirements.

6. An automatic message reply device, using the method as claimed in any one of claims 1 to 5, characterized in that: include: An association module, configured to respond to a target event triggered by a local chat tool, wherein the chat tool is associated with an automatic reply configuration in an input method application in real time in the target event; a dialogue processing module configured to respond to a first message triggered by the chat tool, identify text or image corpus information in the first message using a trained dialogue processing model, parse and obtain chat intent in the corpus information, and generate and output reply content adapted to the chat intent; The information reply module is configured to respond to the reply content generated by the input method application and perform automatic reply via the chat tool using the reply content as a second message.

7. An automatic message reply system, using the method as claimed in any one of claims 1 to 5, characterized in that: include: A remote user terminal, a local user terminal and a cloud, wherein the cloud is respectively connected to the remote user terminal and the local user terminal in communication; the remote user terminal and the local user terminal are both configured with a chat tool and an input method application; the cloud obtains a first message in response to the remote user calling the user input configuration in the input method application in the chat tool of the remote user terminal, feeds back to the local user terminal, and receives a second message replied in the local user terminal and feeds back to the remote user terminal; wherein the local user terminal responds to a target event triggered by a local chat tool, and in the target event, the chat tool associates the automatic reply configuration in the input method application in real time; in response to the first message triggered and received by the chat tool, after recognizing the corpus information of the text or picture in the first message by using the dialogue processing model generated by training, the chat intent in the corpus information is parsed and obtained, and the reply content adapted to the chat intent is generated and output; in response to the reply content generated by the input method application, the reply content is used as the second message to perform an automatic reply through the chat tool.

8. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method according to any one of claims 1 to 5.

9. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images