An internal control learning system and method for acquiring internal control learning content

Abstract

The application discloses an internal control learning system for obtaining internal control learning content, comprising: a terminal side, which is used for generating an internal control learning request containing user behavior text information by processing a collected video stream of a user learning scene and sending the internal control learning request to a network side; and the network side, which is used for calling a related knowledge database to perform matching processing on the user behavior text information according to the internal control learning request containing the user behavior text information from the terminal side, obtaining internal control learning content matched with the user behavior text information, and feeding back the internal control learning content to the terminal side. The terminal side performs processing on the video stream, which comprises: collecting a video stream of a user learning scene; extracting target video time sequence features ft from the video stream; obtaining user behavior text information according to a predetermined correspondence between the target video time sequence features ft and user behaviors; and generating an internal control learning request by analyzing the similarity and duration of each frame of video of the target video time sequence features ft.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video analysis and content recommendation, and particularly relates to an internal control learning system and method for obtaining internal control learning content. BACKGROUND

[0002] With the rapid development of artificial intelligence and Internet of Things technology, intelligent terminal devices are increasingly widely used in education and teaching, industrial detection, security monitoring and other fields. Traditional video signal processing systems usually rely on fixed algorithms or manual analysis, making it difficult to achieve efficient and accurate content recognition and knowledge matching. In particular, in the field of internal control work, how to understand personnel work in real time, collect video signals, extract key features, and combine knowledge bases for intelligent feedback has become a key technical challenge to improve learning efficiency.

[0003] Internal control work is a series of self-restricting, self-adjusting and self-controlling methods and measures established by enterprises or organizations to achieve business objectives, ensure asset safety and integrity, ensure the authenticity and reliability of accounting information, and improve operational efficiency and effectiveness in accordance with laws and regulations.

[0004] However, there is currently a lack of technology for effectively obtaining internal control learning content or knowledge through video analysis. SUMMARY

[0005] The purpose of the present application is to provide an internal control learning system and method for obtaining internal control learning content. The present application obtains relevant internal control knowledge by continuously analyzing the user's office computer screen.

[0006] According to a first aspect of the present application, an internal control learning system for obtaining internal control learning content comprises:

[0007] A terminal side for generating an internal control learning request containing user behavior text information for sending to a network side by processing a collected video stream of a user learning scene;

[0008] A network side for calling a relevant knowledge database to process user behavior text information according to an internal control learning request containing user behavior text information from the terminal side, obtaining internal control learning content matched with the user behavior text information, and feeding back the internal control learning content to the terminal side;

[0009] The terminal side comprises:

[0010] A video signal collector for collecting a video stream of a user learning scene;

[0011] A time sequence feature extraction module for extracting target video time sequence features ft from the video stream;

[0012] a user behavior acquisition module, configured to receive the target video time sequence feature ft and acquire user behavior text information according to a predetermined target video time sequence feature ft and user behavior correspondence relationship;

[0013] a behavior trigger module, configured to generate trigger information for triggering sending of a learning request by analyzing similarity and duration of each frame of video of the target video time sequence feature ft;

[0014] a sending / receiving module, configured to generate and send, according to the trigger information from the behavior trigger module and based on the user behavior text information from the user behavior acquisition module, an internal control learning request containing the user behavior text information to a network side, and receive internal control learning content returned by the network side;

[0015] a display module, configured to display the internal control learning content returned by the network side.

[0016] Preferably, the time sequence feature extraction module is composed of a trained CNN network; the CNN network is trained as a neural network for extracting specific target video time sequence features from a video stream, so as to extract the target video time sequence feature ft from the video stream of the user learning scene.

[0017] Preferably, the behavior trigger module is composed of an LSTM network; the LSTM network generates trigger information when each frame of video of the target video time sequence feature ft is similar and the duration of the similar video frame is greater than a predetermined value.

[0018] Preferably, the user behavior text information includes files related to internal control work, chart names and text information of user processing file and chart behavior.

[0019] Preferably, the network side includes: a matching module, configured to determine a database containing the files and charts according to the file and chart names in the user behavior text information carried by the internal control learning request, and find internal control knowledge matching the user processing file and chart behavior from the database; and a feedback module, configured to feed back the internal control knowledge found by the network side matching module to the terminal side.

[0020] According to a second aspect of the present application, an internal control learning system for acquiring internal control learning content includes:

[0021] a terminal side, configured to generate an internal control learning request containing user behavior video information sent to a network side by processing a video stream of a user learning scene collected by the terminal side;

[0022] On the network side, the internal control learning content is obtained by calling a relevant knowledge database to perform matching processing on the user behavior video information according to the internal control learning request containing the user behavior video information from the terminal side, and the internal control learning content is fed back to the terminal side.

[0023] The terminal side comprises:

[0024] a video signal collector configured to collect a video stream of a user learning scene;

[0025] a time sequence feature extraction module configured to extract target video time sequence features ft from the video stream;

[0026] a behavior trigger module configured to generate trigger information for triggering sending of an internal control learning request by analyzing similarity and duration of each frame of video of the target video time sequence features ft;

[0027] a sending / receiving module configured to generate and send, to the network side, an internal control learning request containing the target video time sequence features ft according to the trigger information from the behavior trigger module and based on the target video time sequence features ft from the time sequence feature extraction module, and receive internal control learning content returned by the network side;

[0028] a display module configured to display the internal control learning content returned by the network side.

[0029] Preferably, the time sequence feature extraction module is composed of a trained CNN network; the CNN network is trained as a neural network for extracting specific target video time sequence features from a video stream, so as to extract the target video time sequence features ft from the video stream of the user learning scene.

[0030] Preferably, the behavior trigger module is composed of an LSTM network; the LSTM network generates the trigger information when each frame of video of the target video time sequence features ft is similar and the duration of the similar video frames is greater than a predetermined value.

[0031] According to a third aspect of the present application, an internal control learning method for obtaining internal control learning content comprises:

[0032] The terminal side generates an internal control learning request containing user behavior text information by processing the collected video stream of the user learning scene and sends the internal control learning request to the network side;

[0033] The network side obtains the internal control learning content matched with the user behavior text information by calling a relevant knowledge database to perform matching processing on the user behavior text information according to the internal control learning request containing the user behavior text information from the terminal side, and feeds back the internal control learning content to the terminal side;

[0034] The terminal side generates an internal control learning request text information containing user behavior information by processing the collected video stream of the user learning scene, and sends the internal control learning request text information to the network side.

[0035] A video signal collector collects a video stream of a user learning scene.

[0036] A time sequence feature extraction module extracts target video time sequence features ft from the video stream.

[0037] A user behavior acquisition module receives the target video time sequence features ft, and acquires user behavior text information according to a predetermined correspondence between the target video time sequence features ft and the user behavior.

[0038] A behavior triggering module generates triggering information for triggering learning request sending by analyzing the similarity and the duration of each frame of video of the target video time sequence features ft.

[0039] A sending / receiving module generates and sends an internal control learning request containing user behavior text information to the network side according to the triggering information from the behavior triggering module and based on the user behavior text information from the user behavior acquisition module, and receives internal control learning content returned by the network side.

[0040] A display module displays the internal control learning content returned by the network side.

[0041] According to a fourth aspect of the present application, an internal control learning method for acquiring internal control learning content comprises:

[0042] A terminal side generates an internal control learning request containing user behavior video information by processing a collected video stream of a user learning scene, and sends the internal control learning request to a network side.

[0043] The network side calls a related knowledge database to perform matching processing on the user behavior video information according to the internal control learning request containing user behavior video information from the terminal side, obtains internal control learning content matched with the user behavior video information, and feeds back the internal control learning content to the terminal side.

[0044] The terminal side generates an internal control learning request containing user behavior video information by processing a collected video stream of a user learning scene, and sends the internal control learning request to a network side, and the generating comprises:

[0045] A video signal collector collects a video stream of a user learning scene.

[0046] A time sequence feature extraction module extracts target video time sequence features ft from the video stream.

[0047] The behavior trigger module generates trigger information for triggering sending of the internal control learning request by analyzing the similarity and duration of each frame of video of the target video time sequence feature ft.

[0048] The sending / receiving module generates and sends an internal control learning request containing the target video time sequence feature ft to the network side according to the trigger information from the behavior trigger module and based on the target video time sequence feature ft from the time sequence feature extraction module, and receives the internal control learning content returned by the network side.

[0049] The display module displays the internal control learning content returned by the network side.

[0050] The present application can understand the working scene of the user in real time and dynamically recommend the most relevant learning content by video signal acquisition and feature analysis, combined with intelligent matching of the database or knowledge brain, to realize the intelligent interaction of "seeing is learning". BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a schematic diagram of a first embodiment of an internal control learning system for obtaining internal control learning content of the present application;

[0052] Figure 2 is a schematic diagram of a second embodiment of an internal control learning system for obtaining internal control learning content of the present application. DETAILED DESCRIPTION

[0053] The technical idea of the present application is to perform image acquisition through a built-in video signal collector. The collected signal is extracted by a feature vector, the main content of the video signal is analyzed in time sequence, the feature vector value is matched with the knowledge of the knowledge brain, and then the relevant learning content is displayed through a display device.

[0054] Figure 1 A first embodiment of an internal control learning system for obtaining internal control learning content of the present application is shown, which comprises: a terminal side for generating an internal control learning request containing user behavior text information for sending to the network side by analyzing the video stream of the collected user learning scene; a network side for calling a relevant knowledge database to match and process the user behavior text information according to the internal control learning request containing the user behavior text information from the terminal side, obtaining the internal control learning content matched with the user behavior text information, and feeding back the internal control learning content to the terminal side.

[0055] The terminal side of the internal control learning system of the application comprises: a video signal collector for collecting a video stream of a user learning scene; a time sequence feature extraction module for extracting target video time sequence features ft from the video stream; a user behavior acquisition module for receiving the target video time sequence features ft and acquiring user behavior text information according to a predetermined correspondence between the target video time sequence features ft and user behaviors; a behavior triggering module for generating trigger information for triggering the sending of a learning request by analyzing the similarity and duration of each frame of video of the target video time sequence features ft; a sending / receiving module for generating and sending an internal control learning request containing user behavior text information to the network side according to the trigger information from the behavior triggering module and based on the user behavior text information from the user behavior acquisition module, and receiving internal control learning content returned by the network side; and a display module for displaying the internal control learning content returned by the network side.

[0056] The user learning scene of the application includes: user handling of OA (office automation system) processes, financial systems, news browsing, watching programs, document writing, etc. The target video time sequence features ft extracted from the video stream by the application are target video time sequence features related to internal control work. The target video time sequence features may be, for example, video time sequence features of user handling of OA processes, video time sequence features of user handling of financial systems, etc. Therefore, the video stream of the user learning scene is processed by the time sequence feature extraction module, and video time sequence features unrelated to internal control work are discarded, such as video time sequence features of news browsing and video time sequence features of watching programs, and the target video time sequence features are acquired.

[0057] Since CNN is suitable for image classification, the time sequence feature extraction module of the application can be composed of a trained CNN network; the CNN network is trained by a labeled training set and is trained into a neural network for extracting specific target video time sequence features from a video stream, so as to extract target video time sequence features ft, such as video time sequence features ft of user handling of OA processes and video time sequence features ft of user handling of financial systems, from all video streams of user learning scenes.

[0058] The application can extract the visual feature vector of each frame from the collected video stream V={I1, I2,..., It}, where It is the t-th frame image, by running a lightweight CNN model (such as MobileNetV3).

[0059] Since the LSTM (Long Short Term Memory) network can establish time sequence dependency relationship, the behavior trigger module of the application can be composed of the LSTM (Long Short Term Memory) network; the LSTM network generates trigger information when each frame of the target video time sequence feature ft is similar and the similar video frame duration is greater than a predetermined value. For example, when each frame of the target video time sequence feature ft is a video frame of the user processing the financial system and the duration is greater than 30 seconds, it indicates that the user is processing the financial system, rather than merely browsing the financial system, and the behavior trigger information that the user is processing the financial system is output.

[0060] The user behavior acquisition module can save a relationship list of the correspondence between all target video time sequence features ft and user behaviors, and by searching the relationship list, the user behavior acquisition module can acquire user behavior text information, such as file and chart names related to internal control work and text information of user processing file and chart behaviors, according to the target video time sequence feature ft.

[0061] The network side of the first embodiment of the internal control learning system of the application comprises: a matching module, configured to determine a database containing a file and a chart according to a file and a chart name in user behavior text information carried by the internal control learning request, and search for internal control knowledge matching the user processing file and chart behaviors; and a feedback module, configured to feed back the internal control knowledge searched by the network side matching module to the terminal side.

[0062] The first embodiment of the application has the following technical effects: 1) the terminal side only sends an internal control learning request to the network side when it is determined that the user's continuous behavior involves internal control work, rather than sending a learning request to the network side in real time, thereby greatly reducing network data transmission and avoiding network congestion caused by excessive data flow; 2) the terminal side sends an internal control learning request to the network side through text information, rather than directly sending a target time sequence feature ft to the network side, thereby reducing data transmission flow.

[0063] Figure 2 A second embodiment of an internal control learning system for acquiring internal control learning content is shown, comprising: a terminal side, configured to generate an internal control learning request containing user behavior video information sent to a network side by processing a collected user learning scene video stream; and a network side, configured to call a related knowledge database to perform matching processing on user behavior video information according to the internal control learning request containing user behavior video information from the terminal side, obtain internal control learning content matched with the user behavior video information, and feed back the internal control learning content to the terminal side.

[0064] In the second embodiment of the internal control learning system, the terminal side comprises: a video signal collector configured to collect a video stream of a user learning scene; a time sequence feature extraction module configured to extract target video time sequence features ft from the video stream; a behavior trigger module configured to generate trigger information for triggering sending of an internal control learning request by analyzing similarity and duration of each frame of video of the target video time sequence features ft; a sending / receiving module configured to generate and send, according to the trigger information from the behavior trigger module and based on the target video time sequence features ft from the time sequence feature extraction module, an internal control learning request containing the target video time sequence features ft to the network side, and receive internal control learning content returned by the network side; and a display module configured to display the internal control learning content returned by the network side.

[0065] In the second embodiment of the internal control learning system, the user learning scene comprises: a user processing OA process, a financial system, news browsing, watching a program, document writing, and the like.

[0066] The time sequence feature extraction module of the second embodiment of the internal control learning system is composed of a trained CNN network; the CNN network is trained as a neural network for extracting specific target video time sequence features from a video stream, so as to extract target video time sequence features ft from the video stream of the user learning scene.

[0067] The behavior trigger module of the second embodiment of the internal control learning system is composed of an LSTM network; the LSTM network generates trigger information when each video frame of the target video time sequence features ft is similar and the duration of the similar video frames is greater than a predetermined value.

[0068] The network side of the second embodiment of the internal control learning system comprises: a matching module configured to perform similarity matching processing on the target video time sequence features ft carried by the internal control learning request and structured data in a knowledge base, and generate adapted internal control knowledge by a deep learning recommendation algorithm; and a feedback module configured to feed back the generated adapted internal control knowledge to the terminal side.

[0069] The main difference between the first embodiment and the second embodiment of the internal control learning system is that the terminal side of the first embodiment generates an internal control learning request containing user behavior text information, and the terminal side of the second embodiment generates an internal control learning request containing user behavior video information. Therefore, the data transmission amount of the first embodiment is lower than that of the second embodiment.

[0070] The application further provides an internal control learning method for obtaining internal control learning content, comprising: a terminal side generates an internal control learning request containing user behavior text information for sending to a network side by processing a collected video stream of a user learning scene; and the network side calls a related knowledge database to perform matching processing on the user behavior text information according to the internal control learning request containing the user behavior text information from the terminal side, obtains internal control learning content matched with the user behavior text information, and feeds back the internal control learning content to the terminal side.

[0071] The terminal side processes the collected video stream of the user learning scene as follows: a video signal collector collects the video stream of the user learning scene; a time sequence feature extraction module extracts target video time sequence features ft from the video stream; a user behavior acquisition module receives the target video time sequence features ft and acquires user behavior text information according to a predetermined correspondence between the target video time sequence features ft and user behaviors; a behavior triggering module generates triggering information for triggering learning request sending by analyzing the similarity and duration of each frame of video of the target video time sequence features ft; a sending / receiving module generates and sends an internal control learning request containing user behavior text information to the network side according to the triggering information from the behavior triggering module and based on the user behavior text information from the user behavior acquisition module, and receives internal control learning content returned by the network side; and a display module displays the internal control learning content returned by the network side.

[0072] The time sequence feature extraction module is composed of a trained CNN network; the CNN network is trained as a neural network for extracting specific target video time sequence features from a video stream, so as to extract target video time sequence features ft from the video stream of the user learning scene.

[0073] The behavior triggering module is composed of an LSTM network; the LSTM network generates triggering information when each frame of video of the target video time sequence features ft is similar and the duration of the similar video frames is greater than a predetermined value.

[0074] The network side processes the internal control learning request from the terminal side as follows: a network side matching module determines a database containing a file and a chart from a database cluster according to a file and a chart name in the user behavior text information carried by the internal control learning request; the network side matching module finds internal control knowledge matched with the user processing file and chart behavior by traversing the database containing the file and the chart; and a network side feedback module feeds back the internal control knowledge found by the network side matching module to the terminal side.

[0075] The application further provides a second embodiment of an internal control learning method for obtaining internal control learning content, comprising: a terminal side processing a collected video stream of a user learning scene to generate an internal control learning request containing user behavior video information and sending the internal control learning request to a network side; and the network side calling a relevant knowledge database to perform matching processing on the user behavior video information according to the internal control learning request containing the user behavior video information from the terminal side, obtaining internal control learning content matched with the user behavior video information, and feeding back the internal control learning content to the terminal side.

[0076] The terminal side processes the collected video stream of the user learning scene as follows: a video signal collector collects the video stream of the user learning scene; a time sequence feature extraction module extracts target video time sequence features ft from the video stream; a behavior trigger module generates trigger information for triggering sending of an internal control learning request by analyzing similarity and duration of each frame of video of the target video time sequence features ft; a sending / receiving module generates and sends an internal control learning request containing the target video time sequence features ft to the network side according to the trigger information from the behavior trigger module and based on the target video time sequence features ft from the time sequence feature extraction module, and receives internal control learning content returned by the network side; and a display module displays the internal control learning content returned by the network side.

[0077] In the second embodiment of the internal control learning method, the time sequence feature extraction module is composed of a trained CNN network; the CNN network is trained as a neural network for extracting specific target video time sequence features from a video stream, so as to extract the target video time sequence features ft from the video stream of the user learning scene.

[0078] In the second embodiment of the internal control learning method, the behavior trigger module is composed of an LSTM network; the LSTM network generates trigger information when each frame of video of the target video time sequence features ft is similar and the duration of the similar video frames is greater than a predetermined value.

[0079] In the second embodiment of the internal control learning method, processing of the internal control learning request by the network side comprises: a network side matching module performing similarity matching of the target video time sequence features ft carried by the internal control learning request with structured data in a knowledge base, and generating adapted internal control knowledge through a deep learning recommendation algorithm; and a network side feedback module feeding back the generated adapted internal control knowledge to the terminal side.

[0080] The application is described below through a financial approval example.

[0081] When a user modifies a purchase contract, a video acquisition module collects a video stream containing the user modifying the purchase contract;

[0082] The time sequence feature extraction module extracts target video time sequence features ft of "amendment clauses" in user modification of procurement contracts from a video stream;

[0083] The behavior triggering module sends a trigger information when it finds that the same clause is modified for three times in succession.

[0084] The user behavior acquisition module sends the target video time sequence features ft of "amendment clauses" to the network side.

[0085] The network side finds the "procurement approval grading system" knowledge and the associated error case "2023 A company's over-level approval event" through the matching processing of the knowledge base.

[0086] The network side generates an approval operation flowchart containing red-labeled irregular operations, and sends the approval operation flowchart and the associated error case, the "compliance approval SOP" system to the terminal side.

[0087] The above technical solutions of the present application can achieve the following technical effects: 1) The video acquisition module can real-time perceive the user learning scene (such as daily login system, document notes, webpage operation); 2) The terminal side sends the internal control learning request to the network only after determining that the user behavior occurs, thereby greatly reducing the information transmission of the network and reducing the interaction delay.

[0088] Although the present application has been described in detail above, the present application is not limited thereto, and those skilled in the art can make various modifications according to the principles of the present application. Therefore, any modification made according to the principles of the present application should be understood as falling within the scope of the present application.

Claims

1. An internal control learning system for acquiring internal control learning content, comprising: On the terminal side, the system processes video streams collected from users' OA workflows, financial systems, news browsing, and program viewing to generate an internal control learning request containing user behavior text information to be sent to the network side. The user behavior text information includes text information about users' actions related to internal control work, such as processing documents and charts. On the network side, it is used to call the relevant knowledge database to match the user behavior text information according to the internal control learning request containing user behavior text information from the terminal side, obtain the internal control learning content that matches the user behavior text information, and feed the internal control learning content back to the terminal side. The terminal side includes: The video signal acquisition module is used to acquire video streams from users' OA process handling, financial system processing, news browsing, and program viewing. The temporal feature extraction module is used to extract target video temporal features ft related to internal control work from the video stream; The user behavior acquisition module is used to receive the target video temporal features ft related to internal control work, and to acquire user behavior text information according to the pre-determined correspondence between the target video temporal features ft related to internal control work and user behavior. The behavior triggering module is used to analyze the similarity and duration of each frame of the target video temporal features ft related to internal control work, and generate trigger information to trigger the sending of learning requests when the duration is greater than a predetermined value. The sending / receiving module is used to generate and send an internal control learning request containing user behavior text information to the network side based on the triggering information from the behavior triggering module and the user behavior text information from the user behavior acquisition module, and to receive the internal control learning content returned by the network side. The display module is used to display the internal control learning content returned by the network side.

2. The internal control learning system according to claim 1, wherein the temporal feature extraction module is composed of a trained CNN network; the CNN network is a neural network trained to extract target video temporal features ft related to internal control work from video streams of users processing OA processes, processing financial systems, browsing news, and watching programs.

3. The internal control learning system according to claim 1 or 2, wherein the behavior triggering module is composed of an LSTM network; the LSTM network generates the triggering information when the duration of video frames similar to each frame of the target video temporal feature ft related to internal control work is greater than a predetermined value.

4. The internal control learning system according to claim 3, wherein the user behavior text information further includes text information of the names of documents and charts related to internal control work.

5. The internal control learning system according to claim 4, wherein the network side comprises: The matching module is used to determine the database containing the file and chart from the database cluster based on the file and chart names in the user behavior text information carried by the internal control learning request, and to search for internal control knowledge that matches the user's behavior of processing the file and chart. The feedback module is used to feed back the internal control knowledge found by the network-side matching module to the terminal side.

6. An internal control learning system for acquiring internal control learning content, comprising: On the terminal side, the system processes video streams collected from users' OA workflows, financial systems, news browsing, and program viewing to generate an internal control learning request containing user behavior video information, which is then sent to the network side. The user behavior video information includes video information of users' actions in processing documents and charts related to internal control work. On the network side, it is used to call the relevant knowledge database to match the user behavior video information according to the internal control learning request containing user behavior video information from the terminal side, obtain the internal control learning content that matches the user behavior video information, and feed the internal control learning content back to the terminal side. The terminal side includes: Video signal acquisition device is used to acquire video streams from users' OA process handling, financial system processing, news browsing, and program viewing. The temporal feature extraction module is used to extract target video temporal features ft related to internal control work from the video stream; The behavior triggering module is used to analyze the similarity and duration of each frame of the target video temporal features ft related to internal control work, and generate triggering information to trigger the sending of internal control learning requests when the duration is greater than a predetermined value. The sending / receiving module is used to generate and send an internal control learning request containing the target video temporal features ft related to internal control work to the network side based on the triggering information from the behavior triggering module and the target video temporal features ft related to internal control work from the temporal feature extraction module, and to receive the internal control learning content returned by the network side. The display module is used to display the internal control learning content returned from the network side.

7. The internal control learning system according to claim 6, wherein the temporal feature extraction module is composed of a trained CNN network; the CNN network is a neural network trained to extract target video temporal features ft related to internal control work from video streams of users processing OA processes, processing financial systems, browsing news, and watching programs.

8. The internal control learning system according to claim 6 or 7, wherein the behavior triggering module is composed of an LSTM network; the LSTM network generates the triggering information when the duration of video frames similar to each frame of the target video temporal feature ft related to internal control work is greater than a predetermined value.

9. The internal control learning system according to claim 8, wherein the user behavior video information further includes video information of the names of documents and charts related to internal control work.

10. An internal control learning method for acquiring internal control learning content, comprising: The terminal side processes the video streams collected from users' OA process handling, financial system processing, news browsing, and program viewing to generate an internal control learning request containing user behavior text information to be sent to the network side. The user behavior text information includes text information about users' actions in handling documents and charts related to internal control work. The network side calls the relevant knowledge database to match the user behavior text information based on the internal control learning request containing user behavior text information from the terminal side, obtains the internal control learning content that matches the user behavior text information, and feeds back the internal control learning content to the terminal side. The terminal side processes the collected video streams from user OA workflows, financial systems, news browsing, and program viewing to generate internal control learning request text information containing user behavior information, which is then sent to the network side. This includes: The video signal acquisition device collects video streams from users' OA workflow processing, financial system processing, news browsing, and program viewing. The temporal feature extraction module extracts target video temporal features ft related to internal control work from the video stream; The user behavior acquisition module receives the target video temporal features ft related to internal control work, and obtains user behavior text information based on the pre-determined correspondence between the target video temporal features ft and user behavior. The behavior triggering module analyzes the similarity and duration of each frame of the target video temporal features ft related to internal control work. When the duration is greater than a predetermined value, it generates triggering information to trigger the sending of learning requests. The sending / receiving module generates and sends an internal control learning request containing user behavior text information to the network side based on the triggering information from the behavior triggering module and the user behavior text information from the user behavior acquisition module, and receives the internal control learning content returned by the network side. The display module shows the internal control learning content returned from the network side.

11. The internal control learning method according to claim 10, wherein the temporal feature extraction module is composed of a trained CNN network; the CNN network is a neural network trained to extract target video temporal features ft related to internal control work from video streams of users processing OA processes, processing financial systems, browsing news, and watching programs.

12. The internal control learning method according to claim 10 or 11, wherein the behavior triggering module is composed of an LSTM network; the LSTM network generates the triggering information when the duration of video frames similar to each frame of the target video temporal feature ft related to internal control work is greater than a predetermined value.

13. The internal control learning method according to claim 12, wherein the user behavior text information further includes text information of the names of documents and charts related to internal control work.

14. An internal control learning method for acquiring internal control learning content, comprising: The terminal side processes the video streams collected from users' OA process handling, financial system processing, news browsing, and program viewing to generate an internal control learning request containing user behavior video information, which is sent to the network side. The user behavior video information includes video information of users' behavior in handling documents and charts related to internal control work. The network side, based on the internal control learning request containing user behavior video information from the terminal side, calls the relevant knowledge database to match the user behavior video information, obtains internal control learning content that matches the user behavior video information, and feeds back the internal control learning content to the terminal side. The terminal side processes video streams collected from user OA workflows, financial systems, news browsing, and program viewing to generate internal control learning requests containing user behavior video information, which are then sent to the network side. The video signal acquisition device collects video streams from users' OA workflow processing, financial system processing, news browsing, and program viewing. The temporal feature extraction module extracts target video temporal features ft related to internal control work from the video stream; The behavior triggering module analyzes the similarity and duration of each frame of the target video temporal features ft related to internal control work. When the duration is greater than a predetermined value, it generates triggering information to trigger the sending of internal control learning requests. The sending / receiving module generates and sends an internal control learning request containing the target video temporal features ft related to internal control work to the network side based on the triggering information from the behavior triggering module and the target video temporal features ft related to internal control work from the temporal feature extraction module, and receives the internal control learning content returned by the network side. The display module shows the internal control learning content returned from the network side.

15. The internal control learning method according to claim 14, wherein the temporal feature extraction module is composed of a trained CNN network; the CNN network is a neural network trained to extract target video temporal features ft related to internal control work from video streams of users processing OA processes, processing financial systems, browsing news, and watching programs.

16. The internal control learning method according to claim 14 or 15, wherein the behavior triggering module is composed of an LSTM network; the LSTM network generates the triggering information when the duration of video frames similar to each frame of the target video temporal feature ft related to internal control work is greater than a predetermined value.

17. The internal control learning method according to claim 16, wherein the user behavior video information further includes video information of the names of documents and charts related to internal control work.

Images