A method, device, storage medium and equipment for automatically filling in information

Abstract

The present application discloses a method, device, storage medium and equipment for automatically filling in information, the method comprising: first obtaining a target document, then inputting the target document into a pre-built text classification model, identifying a preset classification corresponding to the target document; then inputting the target document into a semantic classification model under its corresponding preset classification, identifying the semantic information of the target document, and then automatically recommending the target office software to the target person according to the semantic information, and automatically backfilling the relevant office information when the target person triggers to run the target office software. Since the present application first classifies the target document using a text classification model, and then uses the semantic classification model under the corresponding classification to identify the semantic information of the target document, the target office software can be automatically recommended to the target person and the relevant office information can be backfilled according to the semantic information, thereby reducing the manual operation of the target person and improving the office efficiency of the target person.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method, device, storage medium and equipment for automatically filling in information. Background Art

[0002] With the continuous breakthroughs in artificial intelligence (AI) technology and the increasing popularity of various smart devices, the amount of work people need to handle is gradually increasing. In daily office work, sending messages and emails has become the most natural way of business communication. Recording meeting minutes has also become a common practice in the office. In the past two years, remote meetings have become more frequent. People usually summarize and refine messages, emails, and meeting minutes to clarify work goals and then carry out specific work content, such as opening Excel tables to make schedules, opening Word documents to write copy, arranging meetings to invite people to attend, etc.

[0003] In order to improve people's office efficiency, various AI office platforms have emerged, such as DingTalk and Feishu. However, various AI office platforms currently only support office staff to actively handle various work. For example, when organizing a meeting, office staff need to actively open the meeting invitation interface, enter the meeting title and the invited participants, and then the office platform will only help office staff to check everyone's free time and send meeting invitations. Although these AI office platforms can automatically check the free time of participants and automatically generate meeting records, during this meeting, office staff need to take the initiative to open the meeting invitation interface, enter the meeting title and enter the invited participants, which increases the workload of office staff, and requires staff to judge whether to initiate a meeting. When staff have a lot of work, they may ignore the necessity of the meeting. For the part that office staff fill in by themselves, sometimes the omission of the meeting theme or the omission of the meeting inviter may occur due to subjective mistakes of the staff, resulting in low office efficiency and poor experience for office staff.

[0004] Therefore, how to improve people's office efficiency, reduce errors in the office, and improve people's office experience is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The main purpose of the embodiments of the present application is to provide a method, device, storage medium and equipment for automatically filling in information, which can improve the office efficiency of office personnel and reduce office errors.

[0006] The present application embodiment provides a method for automatically filling in information, including:

[0007] Acquire a target document; the target document is obtained by automatically monitoring the desktop or application of the terminal device;

[0008] Input the target document into a pre-built text classification model to identify the preset classification corresponding to the target document; the preset classification includes chat type, project information type, and work trigger type;

[0009] Inputting the target document into a semantic classification model corresponding to a preset classification to identify and obtain semantic information of the target document;

[0010] According to the semantic information of the target document, the target office software is automatically recommended to the target person, and when the target person triggers to run the target office software, the relevant office information is automatically filled in.

[0011] In a possible implementation, the text classification model is constructed as follows:

[0012] Obtaining a sample document; the sample document is obtained by automatically monitoring the desktop or application of the terminal device;

[0013] Using the sample documents, training an initial text classification model to obtain the text classification model;

[0014] The initial text classification model includes an input layer, an encoding layer, a bidirectional long short-term memory recurrent neural network (LSM) layer, an attention layer, a fully connected layer and an output layer.

[0015] In a possible implementation, inputting the target document into a pre-built text classification model to identify a preset classification corresponding to the target document includes:

[0016] Inputting the word vector of the target document into the encoding layer through the input layer to obtain an encoding vector;

[0017] Input the encoding vector into the Bilstm layer to obtain a context feature vector for each word in the target document;

[0018] Inputting the context feature vector of each word in the target document into the attention layer to obtain a weighted context feature vector;

[0019] The weighted context feature vector is input into the fully connected layer, the target document is classified, and the obtained classification result is output through the output layer.

[0020] In a possible implementation, when the preset category corresponding to the target document is a work trigger type, inputting the target document into a semantic classification model under the corresponding preset category to identify and obtain semantic information of the target document includes:

[0021] Inputting the target document into the first semantic classification model under the work trigger type, identifying the target office software corresponding to the semantic information of the target document;

[0022] and / or, inputting the target document into a second semantic classification model under a work trigger type to identify and obtain project information to which the semantic information of the target document belongs;

[0023] The first semantic classification model and the second semantic classification model are both trained using sample documents belonging to the work trigger classification.

[0024] In a possible implementation, when the preset category corresponding to the target document is a project information type, inputting the target document into a semantic classification model under the corresponding preset category to identify and obtain semantic information of the target document includes:

[0025] Inputting the target document into a third semantic classification model under the project information type to identify the project information to which the semantic information of the target document belongs;

[0026] The third semantic classification model is trained using sample documents belonging to the project information classification.

[0027] In a possible implementation, after inputting the target document into a semantic classification model corresponding to a preset classification and identifying and obtaining semantic information of the target document, the method further includes:

[0028] Extracting key information from the semantic information of the target document;

[0029] Entities and relationships between entities are extracted from the key information to form entity relationship triples, and the entity relationship triples are stored in a database of the project to which the semantic information of the target document belongs.

[0030] In a possible implementation, after automatically recommending the target office software to the target person according to the semantic information of the target document, the method further includes:

[0031] receiving a request for updating the target office software triggered by the target person, and in response to the request, displaying the updated target office software to the target person;

[0032] The first semantic classification model is updated using the updated target office software.

[0033] The present application also provides a device for automatically filling in information, including:

[0034] A first acquisition unit is used to acquire a target document; the target document is obtained by automatically monitoring the desktop or application of the terminal device;

[0035] A first recognition unit is used to input the target document into a pre-built text classification model to identify and obtain a preset classification corresponding to the target document; the preset classification includes a chat type, a project information type, and a work trigger type;

[0036] A second recognition unit is used to input the target document into a semantic classification model corresponding to the preset classification to recognize and obtain semantic information of the target document;

[0037] The recommendation unit is used to automatically recommend the target office software to the target person according to the semantic information of the target document, and automatically fill in the relevant office information when the target person triggers the operation of the target office software.

[0038] In a possible implementation manner, the device further includes:

[0039] A second acquisition unit is used to acquire a sample document; the sample document is obtained by automatically monitoring the desktop or application of the terminal device;

[0040] A training unit, used to train an initial text classification model using the sample document to obtain the text classification model;

[0041] The initial text classification model includes an input layer, an encoding layer, a bidirectional long short-term memory recurrent neural network (LSM) layer, an attention layer, a fully connected layer and an output layer.

[0042] In a possible implementation, the first identification unit includes:

[0043] A first input subunit, configured to input the word vector of the target document into the encoding layer through the input layer to obtain an encoding vector;

[0044] A second input subunit is used to input the encoding vector into the Bilstm layer to obtain a context feature vector of each word in the target document;

[0045] A third input subunit, configured to input the context feature vector of each word in the target document into the attention layer to obtain a weighted context feature vector;

[0046] The fourth input subunit is used to input the weighted context feature vector into the fully connected layer, classify the target document, and output the obtained classification result through the output layer.

[0047] In a possible implementation, when the preset classification corresponding to the target document is a work trigger type, the second recognition unit is specifically configured to:

[0048] Inputting the target document into the first semantic classification model under the work trigger type, identifying the target office software corresponding to the semantic information of the target document;

[0049] and / or, inputting the target document into a second semantic classification model under a work trigger type to identify and obtain project information to which the semantic information of the target document belongs;

[0050] The first semantic classification model and the second semantic classification model are both trained using sample documents belonging to the work trigger classification.

[0051] In a possible implementation, when the preset classification corresponding to the target document is a project information type, the second identification unit is specifically configured to:

[0052] Inputting the target document into a third semantic classification model under the project information type to identify the project information to which the semantic information of the target document belongs;

[0053] The third semantic classification model is trained using sample documents belonging to the project information classification.

[0054] In a possible implementation manner, the device further includes:

[0055] An extraction unit, used to extract key information from the semantic information of the target document;

[0056] The storage unit is used to extract entities and relationships between entities from the key information to form entity relationship triples, and store the entity relationship triples in a database of the project to which the semantic information of the target document belongs.

[0057] In a possible implementation manner, the device further includes:

[0058] A receiving unit, configured to receive a request for updating the target office software triggered by the target person, and in response to the request, display the updated target office software to the target person;

[0059] An updating unit is used to update the first semantic classification model by using the updated target office software.

[0060] In a possible implementation, the office information is attribute information of a project to which the semantic information of the target document belongs; the office information is obtained from a database of the project to which the semantic information of the target document belongs; and the recommendation unit is specifically configured to:

[0061] When the target person triggers the operation of the target office software, the attribute information of the project is obtained from the database of the project to which the semantic information of the target document belongs; and the attribute information of the project is automatically filled back into the corresponding position of the project interface of the target office software.

[0062] The embodiment of the present application also provides a device for automatically filling in information, including: a processor, a memory, and a system bus;

[0063] The processor and the memory are connected via the system bus;

[0064] The memory is used to store one or more programs, and the one or more programs include instructions. When the instructions are executed by the processor, the processor executes any one of the implementations of the above-mentioned method of automatically filling in information.

[0065] An embodiment of the present application also provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes any one of the implementations of the above-mentioned method of automatically filling in information.

[0066] The embodiment of the present application further provides a computer program product. When the computer program product is executed on a terminal device, the terminal device executes any one of the implementation modes of the above-mentioned method for automatically filling in information.

[0067] The embodiment of the present application provides a method, device, storage medium and equipment for automatically filling in information, first obtaining a target document; wherein the target document is obtained by automatically monitoring the desktop or application of a terminal device, then inputting the target document into a pre-built text classification model, identifying and obtaining a preset classification corresponding to the target document; wherein the preset classification includes a chat classification, a project information classification, and a work trigger classification; then, inputting the target document into a semantic classification model under its corresponding preset classification, identifying and obtaining the semantic information of the target document, and then automatically recommending the target office software to the target person according to the semantic information of the target document, and automatically backfilling the relevant office information when the target person triggers the operation of the target office software. It can be seen that since the embodiment of the present application uses a pre-built text classification model to classify the target office documents that the target person allows to collect, and then uses the semantic classification model under the corresponding classification to identify and obtain the semantic information of the target document, it can automatically provide the target person with the required office software and backfill the relevant office information according to the semantic information, thereby reducing the manual operation of the target person, not only improving the office efficiency of the target person, but also reducing errors in the office. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0069] Figure 1 A flowchart of a method for automatically filling in information provided in an embodiment of the present application;

[0070] Figure 2 A schematic diagram of the network structure of a text classification model provided in an embodiment of the present application;

[0071] Figure 3 A schematic diagram of the composition of a device for automatically filling in information provided in an embodiment of the present application. DETAILED DESCRIPTION

[0072] With the continuous breakthroughs in AI technology and the increasing popularity of various smart devices, the workload that people need to handle in daily office work is gradually increasing. For example, office workers need to open Excel tables to make schedules, open Word documents to write copy, arrange meetings to invite people to attend, and so on. Although the specific work content of each type of work is different, the types of work content and the software that needs to be opened when working can be exhaustively listed. For example, office workers determine through communication emails with colleagues that they need to organize a meeting to facilitate the alignment of various information and notify various matters. At this time, office workers need to open the software for meeting invitations, sort out the people they need to invite, summarize the meeting theme, determine the meeting time, and send out meeting invitations. After the meeting, it is also necessary to send meeting minutes to the participants and summarize the to-do items of all parties.

[0073] At present, in order to improve people's office efficiency, various AI office platforms have emerged, such as iFlytek Tingjian and DingTalk. However, the current AI office platforms usually only support office staff to actively handle various work. For example, although these AI office platforms can automatically check the free time of participants and automatically generate meeting records, during this meeting, office staff are required to take the initiative to open the meeting invitation interface, enter the meeting title, and enter the invitees, which increases the manual operation steps of office staff, and requires staff to judge whether to initiate a meeting. When staff have many tasks, they may ignore the necessity of the meeting. For the part that office staff fill in by themselves, sometimes the omission of the meeting theme or the omission of the meeting inviter may occur due to subjective mistakes of the staff, resulting in low office efficiency and poor experience for office staff.

[0074] Therefore, how to improve people's office efficiency, reduce office errors, and improve people's office experience is a technical problem that needs to be solved urgently.

[0075] In order to solve the above defects, the present application provides a method for automatically filling in information, firstly obtaining a target document; wherein the target document is obtained by automatically monitoring the desktop or application of a terminal device, then inputting the target document into a pre-built text classification model, identifying and obtaining a preset classification corresponding to the target document; wherein the preset classification includes a chat classification, a project information classification, and a work trigger classification; then, inputting the target document into a semantic classification model under its corresponding preset classification, identifying and obtaining the semantic information of the target document, and then automatically recommending the target office software to the target person according to the semantic information of the target document, and automatically backfilling the relevant office information when the target person triggers the operation of the target office software. It can be seen that since the embodiment of the present application uses a pre-built text classification model to classify the target office documents that the target person allows to collect, and then uses the semantic classification model under the corresponding classification to identify and obtain the semantic information of the target document, it can automatically provide the target person with the required office software and backfill the relevant office information according to the semantic information, thereby reducing the manual operation of the target person, not only improving the office efficiency of the target person, but also reducing the mistakes in the office.

[0076] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0077] First embodiment

[0078] See also Figure 1 , is a flow chart of a method for automatically filling in information provided in this embodiment, the method comprising the following steps:

[0079] S101: Acquire a target document; wherein the target document is obtained by automatically monitoring the desktop or application of a terminal device.

[0080] It should be noted that, in office communication scenarios, people usually communicate in text in communication software or mailboxes, and people's office content is often hidden in these text texts. Therefore, in order to improve people's office efficiency, this application proposes to analyze the text texts used by people in office communication scenarios, and automatically recommend corresponding office software based on the analysis results, and fill in part of the office content based on the semantic information in the analysis results, so as to improve people's office efficiency, reduce office errors, and enhance office experience.

[0081] Specifically, in this embodiment, any office worker who can automatically fill in information through this embodiment is defined as a target person, and the office documents that the target person allows to be collected obtained by automatically monitoring the desktop or application of the terminal device are defined as target documents. In other words, the acquisition scope of the target document is the scope of authority pre-collected by the target person. For example, if the target person allows, documents in WeChat, mailbox, DingTalk, Feishu, office, etc. can be obtained as target documents to execute subsequent steps S102-S104, realize automatic backfilling of office information and improve the office efficiency of the target person.

[0082] It should be noted that the present embodiment does not limit the language type of the target document, for example, the target document may be a Chinese document or an English document, etc.; the present embodiment does not limit the length of the target document, for example, the target document may be a sentence document or a chapter-level document; the present embodiment does not limit the field of the target document, for example, the target document may belong to the field of science and technology or the field of medicine, etc.

[0083] S102: Input the target document into a pre-built text classification model to identify the preset classification corresponding to the target document; wherein the preset classification includes chat type, project information type, and work trigger type.

[0084] In this embodiment, after the target document is obtained through step S101, in order to improve the office efficiency of the target personnel, realize the automatic backfilling of office information, and reduce office errors, the obtained target document can be further input into the pre-built text classification model, so as to identify the preset classification corresponding to the target document, such as the chat type, project information type, work trigger type, etc., to perform the subsequent step S103. Among them, the chat type document refers to a document that is not very relevant to the work; the project information type document refers to a document that describes the progress of the work project; the work trigger type document refers to a document that directly requires the office personnel to open the office software to create materials (such as tables, etc.).

[0085] For example: Assume that the target document is a message received by the target person in WeChat: "Make a PPT of this week's work progress for me". When the target document appears on the computer desktop and is seen by the target person, the target document can be entered as follows: Figure 2 The text classification model shown recognizes that the preset classification result corresponding to the target document should be a "work trigger" type.

[0086] Or, suppose the target document is an email received by the target person in the mailbox about the project establishment plan and time nodes. When the email is opened and seen by the target person, the target document can be entered into the following format: Figure 2The text classification model shown in the figure identifies that the target document corresponds to the preset classification result of "project information" type. Alternatively, assuming that the target document is a message such as "what to eat for lunch today" received by the target person on WeChat, when the target document appears on the computer desktop and is seen by the target person, the target document can be input as follows Figure 2 The text classification model shown recognizes that the preset classification result corresponding to the target document should be the "chat" type.

[0087] Among them, Figure 2 As shown in the figure, the network structure of the text classification model may include an input layer, an encoding layer (embedding), a bidirectional long short-term memory recurrent neural network (Bi-directional Long Short-Term Memory, referred to as Bilstm) layer, an attention layer, a fully connected layer (fully connected layer) and an output layer.

[0088] Specifically, an optional implementation method is that the specific implementation process of step S102 includes the following steps AD:

[0089] Step A: Input the word vector of the target document into the encoding layer through the input layer to obtain the encoding vector.

[0090] In this implementation, in order to improve the office efficiency of the target personnel, the vector of a preset dimension (such as 400 dimensions) of each word in the target document can be calculated first, and then each word vector is input into the encoding layer through the input layer to obtain the encoding vector. Figure 2 As shown, we can first calculate the 400-dimensional vectors token1, token2…tokenn of each word in the target document, and then input token1, token2…tokenn into the encoding layer embedding through the input layer to obtain the encoding vector.

[0091] Step B: Input the encoded vector into the Bilstm layer to obtain the context feature vector of each word in the target document.

[0092] like Figure 2 As shown, through step A, the word vectors token1, token2...tokenn of each word in the target document are input into the encoding layer embedding through the input layer. After the encoding vector is obtained, the encoding vector can be further input into the Bilstm layer to obtain the context feature vector of each word in the target document, thereby realizing the splicing of bidirectional features.

[0093] Step C: Input the context feature vector of each word in the target document into the attention layer to obtain the weighted context feature vector.

[0094] like Figure 2 As shown in FIG. 1 , after the encoded vector is input into the Bilstm layer through step B to obtain the context feature vector of each word in the target document, the context feature vector of each word in the target document can be further input into the attention layer to obtain the weighted context feature vector, and the weighted summation of each word is realized in this network layer.

[0095] Step D: Input the weighted context feature vector into the fully connected layer, classify the target document, and output the classification result through the output layer.

[0096] like Figure 2 As shown, the context feature vector of each word in the target document is input into the attention layer through step C. After obtaining the weighted context feature vector, the weighted context feature vector can be further input into the fully connected layer to classify the target document, and the softmax is used as the activation function to output the obtained classification result through the output layer. For example, a 3-dimensional vector can be used to represent the classification result, and each dimensional vector corresponds to the probability of representing that the target document belongs to the chat type, the project information type, and the work trigger type. Assuming that the classification result output through the output layer is [0.8, 0.07, 0.13], it can be determined that the probability that the first dimension value 0.8 represents that the target document belongs to the chat type is the highest, that is, the preset classification corresponding to the target document is the chat type.

[0097] Among them, an optional implementation method is that the construction process of the text classification model can specifically include: first obtaining a sample document; wherein the sample document is obtained by automatically monitoring the desktop or application of the terminal device; then using the sample document to train the initial text classification model to obtain a text classification model.

[0098] Specifically, in this implementation, in order to build a text classification model, a lot of preparatory work needs to be done in advance. First, a large number of office documents obtained from the desktop or application of the automatic monitoring terminal device need to be collected as sample documents. Then, the preset classification (i.e., chat type, project information type, or work trigger type) labels to which all sample documents belong are manually marked. Further, the initial text classification model can be trained based on the sample documents and their corresponding preset classification labeling results, thereby generating a text classification model.

[0099] Among them, an optional implementation method is that the initial text classification model may include an input layer, an encoding layer, a Bilstm layer, an attention layer, a fully connected layer and an output layer.

[0100] Specifically, when conducting model training, a sample document can be extracted from the training data as the model input and the preset classification as the output, and multiple rounds of model training can be performed. The preset classification recognition results obtained in each round of training are compared with the corresponding manual annotation results, and the model parameters are updated according to the difference between the two until the preset conditions are met, such as the number of training times meets the preset threshold, or the loss function reaches the minimum value, etc., then the updating of the model parameters is stopped, the training of the text classification model is completed, and a trained text classification model is generated.

[0101] S103: Input the target document into the semantic classification model under its corresponding preset classification to identify and obtain the semantic information of the target document.

[0102] In this embodiment, after the preset classification corresponding to the target document is identified through step S102, in order to improve the office efficiency of the target personnel and reduce office errors, the target document can be further input into the semantic classification model under its corresponding preset classification, so as to identify the semantic information of the target document for executing the subsequent step S104.

[0103] Specifically, an optional implementation method is that when the preset classification corresponding to the target document is a work trigger type, the specific implementation process of this step S103 may include: inputting the target document into the first semantic classification model under the work trigger type, and identifying the target office software corresponding to the semantic information of the target document; and / or inputting the target document into the second semantic classification model under the work trigger type, and identifying the project information to which the semantic information of the target document belongs.

[0104] The network structures of the first semantic classification model and the second semantic classification model may include an input layer, an encoding layer, a Bilstm layer, an attention layer, a fully connected layer, and an output layer. The first semantic classification model and the second semantic classification model are both trained using sample documents belonging to the work trigger classification. The specific construction process is similar to that of the text classification model, except that the input and output data are different. The specific construction process will not be repeated here.

[0105] In this implementation, the classification result of the first semantic classification model is a preset common office purpose, and each corresponds to a corresponding office software. When the target document is input into the first semantic classification model under the work trigger type, the office software corresponding to the semantic information of the target document can be identified (here it is defined as the target office software). For example, a window can pop up next to the target document, and the identified target office software can be recommended to the target person in it, so that the target person can click on the office software he wants to open. For example: the corresponding target office software when making slides can be PowerPoint, keynote, etc., the corresponding target office software when making charts can be Excel, Xmind, MindMaster, etc., the corresponding target office software when contacting and docking can be WeChat, DingTalk, etc., the corresponding target office software when inviting meetings can be iFlytek Hearing, DingTalk, etc., and the corresponding target office software when using memos can be iFlytek Voice Notes, NetEase Cloud Notes, etc. The recognition result of the second semantic classification model can be the specific project information to which the semantic information of the target document belongs, that is, which project (such as voice assistant optimization or multilingual recognition projects) the target document corresponds to, and based on the group chat where the target document is located, the person to whom it is sent, keywords and other features, it can also be determined which project the target document belongs to, so as to execute the subsequent step S104 to locate and automatically fill in the relevant office information.

[0106] Another optional implementation method is that when the preset classification corresponding to the target document is a project information type, the specific implementation process of this step S103 may include: inputting the target document into the third semantic classification model under the project information type, and identifying the project information to which the semantic information of the target document belongs.

[0107] The network structure of the third semantic classification model may also include an input layer, an encoding layer, a Bilstm layer, an attention layer, a fully connected layer, and an output layer. The third semantic classification model is trained using sample documents belonging to the project information classification. The specific construction process is similar to that of the text classification model, except that the input and output data are different. The specific construction process will not be repeated here.

[0108] In this implementation, the classification result of the third semantic classification model is to determine which project the semantic information in the target document belongs to, and then extract key information from the semantic information of the target document. For example, the target document can be input into a semantic understanding model for extracting key information to extract key information such as key time points, key actions, and key people. Then, entities and relationships between entities are extracted from the key information to form entity relationship triples, and the entity relationship triples are stored in the database of the project to which the semantic information of the target document belongs. Specifically, the specific composition form of the entity relationship triples may include (object 1, object 2, relationship). Among them, objects may include time, people, information, etc., and relationships may include causal relationships, tool-user relationships, product-producer relationships, content-container relationships, part-whole relationships, member composition relationships, information and subject relationships, etc. Then, the entity relationship triples composed of the extracted entities and relationships can be stored in the database of the corresponding project to perform the subsequent step S104 to achieve the positioning and automatic backfilling of relevant office information.

[0109] S104: According to the semantic information of the target document, the target office software is automatically recommended to the target person, and when the target person triggers to run the target office software, the relevant office information is automatically filled in.

[0110] In this embodiment, after the target office software and the corresponding project information of the semantic information of the target document are identified through step S103, in order to improve the office efficiency of the target personnel and reduce office errors, the target office software can be automatically recommended to the target personnel based on the semantic information of the target document, and the relevant office information can be automatically backfilled when the target personnel triggers the operation of the target office software.

[0111] Among them, an optional implementation method is that the automatically backfilled related office information refers to the attribute information of the project to which the semantic information of the target document belongs; and the office information is obtained from the database of the project to which the semantic information of the target document belongs. On this basis, the implementation process of "automatically backfilling related office information when the target personnel triggers the operation of the target office software" in this step S104 may include: when the target personnel triggers the operation of the target office software, obtaining the attribute information of the project from the database of the project to which the semantic information of the target document belongs; and automatically backfilling the attribute information of the project to the corresponding position of the target office software running the project interface.

[0112] For example: if the preset classification corresponding to the target document is a work trigger type, then when the target person opens a target office software according to the recommendation, the corresponding database can be queried according to the project to which the target document belongs, and then the attribute information of the project can be obtained according to the various entities and relationships stored in the database, and it will be automatically filled back into the corresponding position of the interface of the target office software running the project. For example, assuming that the preset classification corresponding to the target document is a work trigger type, and the semantic information of the target document (such as semantic information related to meeting invitations) is identified, and the target office software corresponding to the target office software for meetings is iFlytek Hearing, and the project to which it belongs is Project A, then when the target person opens the meeting invitation interface of iFlytek Hearing according to the recommendation, some basic information can be intelligently filled back, such as the "Meeting Inviter" column will automatically be filled with the person entity of Project A participating in the meeting extracted from the database corresponding to Project A, and the "Meeting Topic" column will automatically be filled with the keywords extracted from the database corresponding to Project A, and the related key information will be recommended next to the keyword, so that the target person can click to add it.

[0113] For another example, assuming that the preset classification corresponding to the target document is a work trigger type, and the target office software corresponding to the identified semantic information of the target document is PowerPoint for making slides, then when the target person clicks on PowerPoint according to the recommendation, some keywords and project-related attribute information of the project corresponding to the semantic information of the target document can be intelligently backfilled. For example, these attribute information can be displayed in a pop-up window as a reference for the target person to build a slide structure and fill in the content of the slide accordingly.

[0114] In a possible implementation method of an embodiment of the present application, in order to improve the office experience of the target personnel, after automatically recommending the target office software to the target personnel based on the semantic information of the target document, a request for updating the target office software triggered by the target personnel can be received, and in response to the request, the updated target office software can be displayed to the target personnel, and then the updated target office software can be used to update the first semantic classification model.

[0115] In this implementation, since each person's office habits may be different, for example, some people are used to using the form of slides when making work reports, while others are used to using the form of tables. Therefore, considering the existence of such personalized needs, this application can also receive the target personnel's update request for the target office software, and use the updated target office software to update the first semantic classification model, so that the first semantic classification model can learn the personalized preferences of each target person, so that more accurate target office software can be recommended to the target person in the future. Specifically, after automatically recommending the target office software to the target person according to the semantic information of the target document, a custom update button can be added near the interface of the recommended target office software. When the target person feels that the target office software does not meet his own needs, he can click the custom update button to highlight the target document, and then open the office software he wants to open as the updated target office software. At this time, the updated target office software can be used to update the first semantic classification model, so that the model can strongly learn the custom needs of the target person and improve the subsequent recommendation effect according to the needs of the target person.

[0116] In summary, the present embodiment provides a method for automatically filling in information, firstly obtaining a target document; wherein the target document is obtained by automatically monitoring the desktop or application of a terminal device, then inputting the target document into a pre-built text classification model, identifying and obtaining a preset classification corresponding to the target document; wherein the preset classification includes a chat classification, a project information classification, and a work trigger classification; then, inputting the target document into a semantic classification model under its corresponding preset classification, identifying and obtaining the semantic information of the target document, and then automatically recommending the target office software to the target person according to the semantic information of the target document, and automatically backfilling the relevant office information when the target person triggers the operation of the target office software. It can be seen that since the present application embodiment uses a pre-built text classification model to classify the target office documents that the target person allows to collect, and then uses the semantic classification model under the corresponding classification to identify and obtain the semantic information of the target document, it can automatically provide the target person with the required office software and backfill the relevant office information according to the semantic information, thereby reducing the manual operation of the target person, which not only improves the office efficiency of the target person, but also reduces errors in the office.

[0117] Second embodiment

[0118] This embodiment will introduce a device for automatically filling in information. For related content, please refer to the above method embodiment.

[0119] See also Figure 3 , is a schematic diagram of a composition of a device for automatically filling in information provided in this embodiment, the device 300 includes:

[0120] The first acquisition unit 301 is used to acquire a target document; the target document is obtained by automatically monitoring the desktop or application of the terminal device;

[0121] The first recognition unit 302 is used to input the target document into a pre-built text classification model to identify and obtain a preset classification corresponding to the target document; the preset classification includes a chat type, a project information type, and a work trigger type;

[0122] The second recognition unit 303 is used to input the target document into a semantic classification model corresponding to the preset classification to recognize and obtain semantic information of the target document;

[0123] The recommendation unit 304 is used to automatically recommend the target office software to the target person according to the semantic information of the target document, and automatically fill in the relevant office information when the target person triggers to run the target office software.

[0124] In an implementation of this embodiment, the device further includes:

[0125] A second acquisition unit is used to acquire a sample document; the sample document is obtained by automatically monitoring the desktop or application of the terminal device;

[0126] A training unit, used to train an initial text classification model using the sample document to obtain the text classification model;

[0127] The initial text classification model includes an input layer, an encoding layer, a bidirectional long short-term memory recurrent neural network (LSM) layer, an attention layer, a fully connected layer and an output layer.

[0128] In an implementation of this embodiment, the first identification unit 302 includes:

[0129] A first input subunit, configured to input the word vector of the target document into the encoding layer through the input layer to obtain an encoding vector;

[0130] A second input subunit is used to input the encoding vector into the Bilstm layer to obtain a context feature vector of each word in the target document;

[0131] A third input subunit, configured to input the context feature vector of each word in the target document into the attention layer to obtain a weighted context feature vector;

[0132] The fourth input subunit is used to input the weighted context feature vector into the fully connected layer, classify the target document, and output the obtained classification result through the output layer.

[0133] In an implementation of this embodiment, when the preset classification corresponding to the target document is a work trigger type, the second identification unit 303 is specifically configured to:

[0134] Inputting the target document into the first semantic classification model under the work trigger type, identifying the target office software corresponding to the semantic information of the target document;

[0135] and / or, inputting the target document into a second semantic classification model under a work trigger type to identify and obtain project information to which the semantic information of the target document belongs;

[0136] The first semantic classification model and the second semantic classification model are both trained using sample documents belonging to the work trigger classification.

[0137] In an implementation of this embodiment, when the preset classification corresponding to the target document is a project information type, the second identification unit 303 is specifically configured to:

[0138] Inputting the target document into a third semantic classification model under the project information type to identify the project information to which the semantic information of the target document belongs;

[0139] The third semantic classification model is trained using sample documents belonging to the project information classification.

[0140] In an implementation of this embodiment, the device further includes:

[0141] An extraction unit, used to extract key information from the semantic information of the target document;

[0142] The storage unit is used to extract entities and relationships between entities from the key information to form entity relationship triples, and store the entity relationship triples in a database of the project to which the semantic information of the target document belongs.

[0143] In an implementation of this embodiment, the device further includes:

[0144] A receiving unit, configured to receive a request for updating the target office software triggered by the target person, and in response to the request, display the updated target office software to the target person;

[0145] An updating unit is used to update the first semantic classification model by using the updated target office software.

[0146] In an implementation of this embodiment, the office information is attribute information of the project to which the semantic information of the target document belongs; the office information is obtained from a database of the project to which the semantic information of the target document belongs; the recommendation unit 304 is specifically used to:

[0147] When the target person triggers the operation of the target office software, the attribute information of the project is obtained from the database of the project to which the semantic information of the target document belongs; and the attribute information of the project is automatically filled back into the corresponding position of the project interface of the target office software.

[0148] Furthermore, an embodiment of the present application also provides a device for automatically filling in information, including: a processor, a memory, and a system bus;

[0149] The processor and the memory are connected via the system bus;

[0150] The memory is used to store one or more programs, and the one or more programs include instructions. When the instructions are executed by the processor, the processor executes any one of the implementation methods of the above-mentioned method of automatically filling in information.

[0151] Furthermore, an embodiment of the present application also provides a computer-readable storage medium, in which instructions are stored. When the instructions are executed on a terminal device, the terminal device executes any one of the implementation methods of the above-mentioned method of automatically filling in information.

[0152] Furthermore, an embodiment of the present application also provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute any one of the implementation methods of the above-mentioned method for automatically filling in information.

[0153] It can be known from the description of the above implementation mode that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a storage medium such as ROM / RAM, a disk, an optical disk, etc., including several instructions for enabling a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in the various embodiments of the present application or certain parts of the embodiments.

[0154] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0155] It should also be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0156] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for automatically filling in information, characterized in that: include: Get the target document; The target document is obtained by automatically monitoring the desktop or application of the terminal device; Input the target document into a pre-built text classification model to identify the preset classification corresponding to the target document; the preset classification includes chat type, project information type, and work trigger type; Inputting the target document into a semantic classification model corresponding to the preset classification to identify and obtain semantic information of the target document; According to the semantic information of the target document, the target office software is automatically recommended to the target person, and when the target person triggers to run the target office software, the relevant office information is automatically filled in; The text classification model is constructed as follows: Obtaining a sample document; the sample document is obtained by automatically monitoring the desktop or application of the terminal device; Using the sample documents, training an initial text classification model to obtain the text classification model; The initial text classification model includes an input layer, an encoding layer, a bidirectional long short-term memory recurrent neural network Bilstm layer, an attention layer, a fully connected layer and an output layer; The step of inputting the target document into a pre-built text classification model to identify a preset classification corresponding to the target document includes: Inputting the word vector of the target document into the encoding layer through the input layer to obtain an encoding vector; Input the encoding vector into the Bilstm layer to obtain a context feature vector for each word in the target document; Inputting the context feature vector of each word in the target document into the attention layer to obtain a weighted context feature vector; The weighted context feature vector is input into the fully connected layer, the target document is classified, and the obtained classification result is output through the output layer.

2. The method according to claim 1, characterized in that When the preset classification corresponding to the target document is a work trigger type, inputting the target document into a semantic classification model under the corresponding preset classification to identify and obtain semantic information of the target document includes: Inputting the target document into the first semantic classification model under the work trigger type, identifying the target office software corresponding to the semantic information of the target document; and / or, inputting the target document into a second semantic classification model under a work trigger type to identify and obtain project information to which the semantic information of the target document belongs; The first semantic classification model and the second semantic classification model are both trained using sample documents belonging to the work trigger classification.

3. The method according to claim 1, characterized in that When the preset classification corresponding to the target document is a project information type, inputting the target document into a semantic classification model under the corresponding preset classification to identify and obtain semantic information of the target document includes: Inputting the target document into a third semantic classification model under the project information type to identify the project information to which the semantic information of the target document belongs; The third semantic classification model is trained using sample documents belonging to the project information classification.

4. The method according to claim 3, characterized in that After inputting the target document into the semantic classification model under the corresponding preset classification to identify the semantic information of the target document, the method further includes: Extracting key information from the semantic information of the target document; Entities and relationships between entities are extracted from the key information to form entity relationship triples, and the entity relationship triples are stored in a database of the project to which the semantic information of the target document belongs.

5. The method according to claim 2, characterized in that: After automatically recommending the target office software to the target person according to the semantic information of the target document, the method further includes: receiving a request for updating the target office software triggered by the target person, and in response to the request, displaying the updated target office software to the target person; The first semantic classification model is updated using the updated target office software.

6. The method according to any one of claims 2 to 5, characterized in that: The office information is the attribute information of the project to which the semantic information of the target document belongs; the office information is obtained from the database of the project to which the semantic information of the target document belongs; When the target personnel triggers to run the target office software, automatically filling in relevant office information includes: When the target person triggers the running of the target office software, the attribute information of the project is obtained from the database of the project to which the semantic information of the target document belongs; The attribute information of the project is automatically filled back into the corresponding position of the interface of the target office software running the project.

7. A device for automatically filling in information, characterized in that: include: A first acquisition unit, used for acquiring a target document; The target document is obtained by automatically monitoring the desktop or application of the terminal device; A first recognition unit is used to input the target document into a pre-built text classification model to identify and obtain a preset classification corresponding to the target document; the preset classification includes a chat type, a project information type, and a work trigger type; A second recognition unit is used to input the target document into a semantic classification model corresponding to the preset classification to recognize and obtain semantic information of the target document; A recommendation unit, configured to automatically recommend a target office software to a target person according to the semantic information of the target document, and automatically fill in relevant office information when the target person triggers the operation of the target office software; The device also includes: A second acquisition unit is used to acquire a sample document; the sample document is obtained by automatically monitoring the desktop or application of the terminal device; A training unit, used to train an initial text classification model using the sample document to obtain the text classification model; The initial text classification model includes an input layer, an encoding layer, a bidirectional long short-term memory recurrent neural network Bilstm layer, an attention layer, a fully connected layer and an output layer; The first identification unit comprises: A first input subunit, configured to input the word vector of the target document into the encoding layer through the input layer to obtain an encoding vector; A second input subunit is used to input the encoding vector into the Bilstm layer to obtain a context feature vector of each word in the target document; A third input subunit, configured to input the context feature vector of each word in the target document into the attention layer to obtain a weighted context feature vector; The fourth input subunit is used to input the weighted context feature vector into the fully connected layer, classify the target document, and output the obtained classification result through the output layer.

8. A device for automatically filling in information, characterized in that: include: Processor, memory, system bus; The processor and the memory are connected via the system bus; The memory is used to store one or more programs, wherein the one or more programs include instructions, and when the instructions are executed by the processor, the processor executes the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a terminal device, the terminal device executes the method according to any one of claims 1 to 6.

Images