TW202329000A - Data processing system and method thereof - Google Patents

Abstract

The present disclosure provides a data processing system and method thereof. The data processing system includes an arranging-inference model and a conference record unit. The arranging-inference model is used to receive an initial presentation file and a data set, and execute an arranging-inference operation according to the initial presentation file and the data set to generate an arranged presentation file. The conference record unit is used to record play-time of the arranged presentation file, and record voices of a plurality of participants of the conference so as to count the number of words of speeches by the participants. Furthermore, it is determined whether to re-train the arranging-inference model based on the play-time of the arranged presentation file and the number of words of speeches by the participants.

Description

Data processing system and method thereof

This disclosure relates to a data processing system and method thereof, in particular to a processing system and method for editing and adjusting meeting briefing materials.

In a meeting, many participants have different backgrounds and experiences, and participants with different backgrounds and experiences are interested in different topics. If the content and layout of the presentation file of the meeting are not oriented to the interests of the meeting participants, most of the meeting participants may lose interest.

When a presentation file that does not meet the interests of the participants is played in the meeting, the situation that often occurs is: the enthusiasm of the participants to speak is reduced, or the participants urge the presenter to speed up the playback speed of the presentation file (the period can end as soon as possible Meeting). Therefore, the briefing material carefully prepared by the speaker before the meeting cannot focus on the correct direction in the actual meeting, which will greatly reduce the effect of the meeting.

In response to the above technical problems, technical personnel in related industries in this technical field are committed to intelligently adjusting and editing the briefing files, so that the edited briefing files can be oriented to the interests of the participants, so as to enhance the participation of the participants in the meeting. degree.

In the technical solution disclosed in this disclosure, the initial briefing file of the meeting is adjusted and arranged by the arrangement reasoning model, and the playback time of the edited briefing file and the number of speech words of the participants are recorded during the meeting, so as to judge whether The orchestration inference model needs to be optimized. Accordingly, the optimized arrangement reasoning model can adjust and arrange the briefing file to be the most suitable content for the meeting participants.

According to one aspect of the present disclosure, a data processing system is provided, and the data processing system includes an arrangement reasoning model and a conference minutes unit. The orchestration reasoning model is used to receive the initial briefing file and the data set, and execute the orchestration reasoning operation according to the initial briefing file and the data set to generate the arranged briefing file. The meeting record unit is used to record the playback time of the briefing file after editing, and record the voices of multiple participants in the conference and count the number of speeches of the participants, and according to the playback time of the briefing file after editing and the speeches of the participants The number of words determines whether to retrain the orchestration reasoning model. Wherein, the data set is at least related to the type and style of the initial briefing file and the background and experience of the participants, and the data set is used to initially train the arrangement reasoning model.

According to another aspect of the present disclosure, a data processing method is provided, and the data processing method includes the following steps. An initial presentation file and a data set are received, wherein the data set is at least related to the type and style of the initial presentation file. Preliminary training and orchestration of inference models based on data sets. The choreography reasoning operation is performed by the choreography reasoning model according to the initial briefing file and the data set to generate the choreography briefing file. Record the playback time of the presentation file after editing. Record the voices of multiple conference participants and count the number of words spoken by the conference participants, wherein the collection of data is at least related to the background and experience of the conference participants. And, it is judged whether to retrain the arrangement reasoning model according to the playback time of the arranged briefing file and the number of words spoken by the participants.

According to the above-mentioned technical solution disclosed in this disclosure, before the meeting, according to the type and style of the briefing file, the playing time of the briefing file, the background experience of the participants, and the degree of participation of the participants (the number of speech words, body movements or expressions) ) preliminarily trains the orchestration reasoning model, based on which the initial briefing file is arranged and adjusted. When the meeting is actually going on, the meeting recording unit can actually analyze the playing time of the arranged briefing file and the actual participation degree of the meeting participants, so as to optimize the arrangement reasoning model.

Other aspects and advantages of this disclosure can be seen by reading the following drawings, detailed description and claims.

The technical terms in this specification refer to the customary terms in this technical field. If some terms are explained or defined in this specification, the explanations or definitions of these terms shall prevail. Each embodiment of the disclosure has one or more technical features. On the premise of possible implementation, those skilled in the art may selectively implement some or all of the technical features in any embodiment, or selectively combine some or all of the technical features in these embodiments.

FIG. 1 is a block diagram of a data processing system 1000 according to an embodiment of the present disclosure. Please refer to FIG. 1 , the data processing system 1000 may include a data set processing unit 100 , an orchestration inference model 200 , a conference record unit 300 and a model optimization unit 400 . The data processing system 1000 can intelligently adjust and arrange the initial presentation file B0 of the meeting, so that the initial presentation file B0 meets the interest orientation of the meeting participants.

For the "windows" platform, the initial presentation file B0 can be, for example, a ".ppt" file or a ".pptx" file generated by the "power point" application program of Microsoft Inc., and the user of the data processing system 1000 For example, the speaker of a conference. Before the meeting, the user can pre-create an initial presentation file B0 and provide the initial presentation file B0 to the data collection processing unit 100 of the data processing system 1000 .

The data set processing unit 100 can classify, name or label the initial presentation file B0 according to the type, style and style of the content of the initial presentation file B0. In one example, the data set processing unit 100 can convert each page of the initial presentation file B0 into a picture file (for example, a ".jpg" file, a ".bmp" file, a ".png" file, etc.), and Analyze the type, style and style of the initial briefing file B0 according to the features in the graphic file. 2A-2F are schematic diagrams of the data collection processing unit 100 classifying, naming or labeling the initial briefing file B0. Please refer to the example in FIG. 2A first. When the corresponding image file converted from the initial presentation file B0 has a large number of features F11 of "medium and large boxes", the data collection processing unit 100 can classify the initial presentation file B0 and The type marked as "overview" or "high-level architecture"; accordingly, the classification label data D1 of the initial presentation file B0 can be set as "overview" or "high-level architecture". In addition, in the example shown in FIG. 2B, when the initial presentation file B0 has a large number of "small frame" features F12 in the corresponding image file, the initial presentation file B0 can be classified and marked as "system architecture (system architecture) )" type; accordingly, the classification label data D1 of the initial presentation file B0 can be set as "system architecture". In the example shown in Figure 2C, when the initial presentation file B0 has a larger number of "short text" (and larger font) features F13 in the corresponding image file, the classification label data D1 of the initial presentation file B0 can be set Either "topic" or "conclusion". In the example shown in Figure 2D, when the corresponding image file of the initial presentation file B0 has the feature F14 of a large number of dense "lined text" (and the font size is small), the classification of the initial presentation file B0 can be set The marked data D1 is "detailed description". In the example shown in FIG. 2E, when the corresponding graphic file of the initial presentation file B0 has the feature F15 of "table", the classification label data D1 of the initial presentation file B0 can be set as "comparison data".

In another example, the type of the initial briefing file B0 can also be analyzed from the perspective of dynamic images. As shown in FIG. 2F , when the initial presentation file B0 displays continuously dynamically changing pictures F16 (for example, animation or video), the classification label data D1 of the initial presentation file B0 can be set as "demo". In yet another example, the data collection processing unit 100 may also have a textual analysis capability, and may perform textual analysis on the text content of the initial briefing file B0. When the context of the text in the initial presentation file B0 is a broad description, the classification label data D1 of the initial presentation file B0 can be set as "summary introduction" or "high-level structure". In contrast, when the meaning of the text of the initial presentation file B0 is detailed description, the classification label data D1 of the initial presentation file B0 can be set as "system structure" or "detailed description".

The above is to analyze the "visual attribute" of the initial presentation file B0 (visual perception of graphics or text). On the other hand, the data set processing unit 100 can also analyze the "time attribute" of the initial presentation file B0. Please refer to FIG. 2G , which shows a schematic diagram of analyzing the time attribute of the initial briefing file B0 by the data collection processing unit 100 . For example, the individual playing time of each page of the initial presentation file B0, the total playing time of all pages of the initial presentation file B0, etc. can be analyzed. Accordingly, the data collection processing unit 100 can, for example, set the playing time data D2 of the initial briefing file B0 as "15 seconds/first page p1, 10 seconds/second page p2, ..., 10 seconds/last page p30" or " 30 minutes and 20 seconds/all pages p1~p30", etc.

In addition to analyzing the initial presentation file B0, the data collection processing unit 100 can also analyze the participants who are invited to attend the meeting. Figure 3 is a schematic diagram of the data collection processing unit 100 analyzing the conference participants; for example, there are sixty conference participants, which are conference participants A1~A60, and Figure 3 is based on one of the conference participants A1 as an example. illustrate.

Please refer to FIG. 3 , before the meeting, the user can send a meeting notice N0 to notify the invited participant A1. The meeting notice N0 is, for example, an email, a message of a communication software, and the like. The data collection processing unit 100 can analyze the information recorded in the meeting notice N0 to obtain the background, department and title N01 of the meeting participant A1, the project plan and work experience N02 of the meeting participant A1, and the key performance indicators of the meeting participant A1 (KPI) N03, or information N04 input by participant A1, etc. The data collection processing unit 100 can further analyze the "background attribute" of the participant A1 according to the above information. In one example, when participant A1's background, department and title N01 is "R&D Engineering Department - Analog IC Design Engineer", and the project plan and work experience N02 is "Design Verification of Bluetooth Wireless Headphone Receiver Chip" ”, the data collection processing unit 100 can set the background attribute data D3 of the participant A1 as “hardware R&D engineer”. In another example, when participant A1's background, department and title N01 is "Marketing Business Department - Product Marketing Manager", and the project plan and work experience N02 is "Overseas marketing of Bluetooth wireless headset products" , the data collection processing unit 100 can set the background attribute data D3 of the participant A1 as "marketing business supervisor".

On the other hand, please refer to Figures 1 and 3 at the same time, the data collection processing unit 100 can also retrieve the meeting history data H0 of the meeting participant A1 from the database 2000 according to the meeting notice N0, and analyze the meeting participants according to the meeting history data H0 A1's historical (ie past time) "participation" in the meeting (ie, the degree of concentration that the participant A1 devotes to the meeting or the degree of enthusiasm in making a speech, etc.). The database 2000 can be built in the data processing system 1000, or can be set independently from the data processing system 1000 (the database 2000 in the embodiment in FIG. 1 is set independently from the data processing system 1000). In one example, the conference history data H0 includes statistical information on the number of speech words A1-1 of the conference participant A1 in the conference. When the number of speech words A1-1 is large, it means that participant A1 has a higher interest in the content of the meeting briefing and is more active in commenting. It can be judged that participant A1 has always had a higher degree of "participation" in the meeting. For example, when the average number of speech words A1-1 of participant A1 in each past meeting is greater than 5000 words, the participation level data D4 of participant A1 can be set as the highest full-level point "100 points". In contrast, when the average number of speech words A1-1 of participant A1 is relatively small, it can be judged that his "participation level" is low; for example, when the average number of speech words A1-1 of participant A1 is less than 20 words, The participation level data D4 of the conference participant A1 can be set as the minimum point "0 point".

In another example, the meeting history data H0 may include video images of the participant A1 in previous meetings. The data collection processing unit 100 can analyze the body movements A1-2 or facial expressions A1-3 of the meeting participant A1 in the meeting from the recorded images to determine their "participation degree" in the meeting. For example, when the body movement A1-2 of the meeting participant A1 is dozing off, looking around, or his facial expression A1-3 is impatience and lack of concentration, the data processing system 1000 can set the participation level data of the meeting participant A1 D4 is the lower number of points "5 points".

From the above, the data collection processing unit 100 also analyzes the meeting notice N0 and the meeting history data H0 of other meeting participants A2-A60 to obtain the background attribute data D3 and participation degree data D4 of other meeting participants A2-A60.

Please refer to FIG. 1 again, the data collection processing unit 100 can collect the classification label data D1, play time data D2 of the initial briefing file B0, and the background attribute data D3 and participation degree data D4 of the participants A1~A60 into a data collection DS, and transmit the data set DS to the orchestration reasoning model 200 . Before the meeting, pre-training can be performed on the orchestration reasoning model 200 by using a neural network (NN) algorithm according to the data set DS. An example of the preliminary training of the orchestration reasoning model 200 is: when the content of the initial briefing file B0 is related to hardware R&D technology, and the background attribute data D3 of most of the participants A1~A60 are "hardware R&D engineers" and participate When the degree data D4 is a relatively high point "90 points", it means that most of the participants A1-A60 are highly interested in the content of the meeting briefing that describes the details of hardware R&D. Accordingly, the orchestration reasoning model 200 can be trained to set a higher weight value for the page where the classification label data D1 is "system architecture", "detailed description" or "comparison data", so as to adjust the orchestration reasoning model training to focus on Detailed description and comparison of hardware architectures.

Another example of the preliminary training of the orchestration reasoning model 200 is: when the initial briefing file B0 is related to hardware research and development technology, and the background attribute data D3 of most of the participants A1~A60 is "marketing business executive" and the participation level data When D4 is "20 o'clock", it means that most of the meeting participants A1~A60 may be very indifferent to the briefing content of the detailed description of hardware development, and the reasoning model 200 can be trained to classify and mark the data D1 as "Summary Introduction" , "Title" or "Conclusion" pages are set to a higher weight value, and the playback time of the page is shortened to adjust the choreography inference model training to focus on a simple overview of the hardware and shorten the overall presentation time.

FIG. 4 is a schematic diagram of the operation of the orchestration reasoning model 200 according to an embodiment of the present disclosure. Please refer to FIG. 4 , after the choreography reasoning model 200 completes the preliminary training, the initial briefing file B0 can be input into the choreography reasoning model 200 to execute the choreography reasoning operation 250 to generate the choreography briefing file B1.

In the step S251 of the layout reasoning operation 250, it can be determined whether any page in the initial briefing file B0 needs to be adjusted or corrected. For example, the last page p30 of the initial briefing file B0 is a summary comment, but the inference result of the arrangement reasoning model 200 shows that the last page p30 has too many detailed text descriptions or detailed diagrams and is not suitable for summary comments, so the judgment result of step S251 For: At least the last page p30 needs to be adjusted and corrected. Then, execute step S252: list the pages (including at least the last page p30) in the initial briefing file B0 that need to be adjusted and amended, and suggest the user to adjust and amend the contents of these pages.

On the other hand, if the judgment result of step S251 is "No", it means that the reasoning result analyzes that the content of each page of the initial presentation file B0 does not need to be adjusted and corrected, and then step S253 is executed: automatically arrange and adjust all pages of the initial presentation file B0 The order of the pages (in other words, the content of each page is not adjusted, only the order of the pages). For example, the order can be adjusted according to the weight value of each page; when the weight value is larger, it means that the participants A1-A60 have higher interest, and the page with the larger weight value can be adjusted to a higher page order.

Then, when the meeting is actually going on, the user actually plays the programmed briefing file B1 in the meeting, and uses the meeting recording unit 300 to actually monitor and record the progress of the meeting (including the actual playback status of the programmed briefing file B1 and the meeting participant A1 ~A60 actual participation status). Please refer to FIG. 5A first, which shows a block diagram of a conference recording unit 300 according to an embodiment of the present disclosure. The conference recording unit 300 includes a presentation page number recording device 310 , a sound receiving device 320 , a presentation playing time recording device 330 , and an image capture device 340 . Wherein, the presentation page number recording device 310 can record the page number of the current playing page of the edited presentation file B1 (for example, the page number of the currently playing page p5 is "the fifth page"). The presentation page number recording device 310 can also judge whether the presentation file B1 has been played after the whole arrangement (for example, the last page p30 has been played, indicating that the presentation file B1 has been played after the entire arrangement), at this time, the presentation page number recording device 310 can Send the control signal S1 to the radio device 320 and the presentation time recording device 330 . The control signal S1 is the playing end signal.

Next, please refer to FIG. 5B , which shows a schematic diagram of the operation of the sound receiving device 320 and the image capturing device 340 according to an embodiment of the present disclosure. The image capturing device 340 can capture images of a plurality of people at the meeting site and/or remote connections and perform image recognition, thereby identifying the participants A1~A60 who are actually invited to attend the meeting, and recording the participants A1~A60 Image Img of A60 (ie, video recording is performed). On the other hand, the audio receiving device 320 can receive and record the voice Vocs of the real conference participants A1 - A60 identified from the images (that is, record them). The audio receiving device 320 can further include a voice-to-text conversion device 321 to convert the voice Voc of the participants A1~A60 into text and calculate the number of words, so as to accumulate the speech words of the participants A1~A60 (to participate in the meeting Take speaker A1 as an example, accumulatively count the number of words A1-1).

When the sound receiving device 320 receives the control signal S1 (indicating that the entire edited briefing file B1 has been played), the voice-to-text conversion device 321 can analyze whether the accumulative number of words spoken by participants A1~A60 during the entire meeting is equal to zero, Based on this, it is judged whether the conference participants A1~A60 are in the state of "zero speech" (that is, they did not express any words during the entire conference process), and the number of the conference participants A1~A60 in the state of "zero speech" can be counted. When the briefing file B1 after the entire arrangement has been played and the number of participants A1~A60 in the state of "zero speech" exceeds half of the total number (more than half of thirty people), the voice-text conversion device 321 can transmit the control The signal S2 is sent to the presentation time recording device 330 . The control signal S2 indicates that more than half of the participants A1~A60 are in the state of "zero speech" during the entire meeting, indicating that more than half of the people are not interested in the content of the briefing file B1 after the arrangement, and the judgment needs to be based on the arrangement reasoning model 200 Optimizing (ie, retraining orchestration inference model 200 ) is performed.

Next, please refer to FIG. 5C , which shows a schematic diagram of the coordinated operation of the presentation playback time recording device 330 and the presentation page number recording device 310 . The presentation playback time recording device 330 judges according to the control signal S1 that the edited briefing file B1 has not finished playing and according to the control signal S2 judges that not all participants A1~A60 are in the state of "zero speech", indicating that the edited briefing file B1 is still playing and there are still participants in the meeting who have made speeches, then the presentation playback time recording device 330 further analyzes whether the briefing file B1 after editing has at least two consecutive pages whose respective playback times are less than the lower limit, and through the briefing page number recording device 310 Analyze the number of pages that mark the two consecutive pages above. For example, it can be analyzed that the playing time of page p21 (its page number is "page 21") is four seconds and the playing time of the next consecutive page p22 (its page number is "page 22") is Three seconds, the playback time of both is less than the lower limit "five seconds", which means that participants A1~A60 may not be interested in the content of the previous page p20 (the page number is "twentieth page"), so they urge The presenter (the user of the data processing system 1000) speeds up playing the subsequent pages p21, p22 to shorten the playing time. Alternatively, participants A1-A60 may not get satisfactory answers when they ask questions about the contents of the previous page p20, and thus urge the presenter to play the subsequent pages p21 and p22 faster to shorten the playing time.

The example where the playback time of two consecutive pages is less than the lower limit is common in the participants A1~A60 including high-ranking executives who are not interested in (or not very satisfied with) the briefing files of their subordinates. Urges to speed up playback. If the presentation playing time recording device 330 analyzes that the playing time of at least two consecutive pages is less than the lower limit, a control signal S3 is generated. The control signal S3 indicates that the conference participants A1-A60 urge to speed up the playback, so that the playback time of the edited briefing file B1 is shortened, and it is judged that the arrangement reasoning model 200 needs to be optimized.

In another example, the presentation page number recording device 310 can cooperate with the speech-to-text conversion device 321 to analyze those pages in the briefing file B1 after editing, when the number of words spoken by the conference participants A1-A60 is zero, and this These pages are not of interest to conference participants A1-A60. The briefing page number recording device 310 can mark the page numbers of these pages, and the layout reasoning model 200 can consider the characteristics of these pages when optimizing, and can make corrections and adjustments for similar characteristics when optimizing the model.

According to the above examples in FIGS. 5A-5C , the meeting recording unit 300 can record and analyze the progress of the meeting, so as to judge whether the edited briefing file B1 meets the interests of the meeting participants A1-A60 when the meeting is actually in progress. When it is judged that the compiled briefing file B1 is not of interest, the conference recording unit 300 may transmit control signals S2 and S3 to enable the model optimization unit 400 to optimize the compiled reasoning model 200 . That is, the control signals S2 and S3 are trigger signals for activating the model optimization unit 400 .

In another example, the tone of the conference participants A1~A60 can also be analyzed by the audio receiving device 320 according to the voice Voc, or the facial expressions of the conference participants A1~A60 can be analyzed by the image capture device 340 according to the image Img. It is judged whether the conference participants A1 - A60 are not interested in the contents of the arranged briefing file B1 , and the model optimization unit 400 needs to be activated to optimize the arranged reasoning model 200 .

In yet another example, the presentation page number recording device 310 can also be used to analyze the number of pages that have been played so far (for example, fifteen pages p1~p15 have been played), and cooperate with the presentation playback time recording device 330 to analyze the number of pages that have been played. The total playing time of the pages (for example, the total playing time of the fifteen pages p1-p15 is one hour). If the expected time of the meeting is three hours and the briefing file B1 has 30 pages in total after editing, and fifteen pages have been played so far, it means that the briefing file B1 after editing has been played by half, but the total number of fifteen pages The playback time only accounts for one-third of the expected time of the meeting, which means that the playback speed of the edited briefing file B1 is too fast; participants A1~A60 may not be interested in the content of the edited briefing file B1 and urge to speed up the playback speed, then It is judged that it is necessary to start the model optimization unit 400 to perform optimization of the orchestration inference model 200 .

Moreover, the conference recording unit 300 can analyze the voice Voc and image Img of the participants A1-A60 in this meeting and whether the participants A1-A60 are in the state of "zero speech" (take the participant A1 as an example, The analysis results of the number of speech words A1-1), etc., are collected as meeting history data H1 and stored in the database 2000 .

From the above, it can be judged whether to trigger or activate the model optimization unit 400 to perform optimization according to the control signals S2 and S3; wherein, the model optimization unit 400 can include a status controller and an observation agent mechanism (observation agent) to perform optimization. Please refer to FIG. 6 , which shows a flow chart of the operation of the model optimization unit 400 according to an embodiment of the present disclosure. Firstly, the state controller is executed in step S601; the state controller can detect whether there are two consecutive pages p(n), p(n+1) whose playing time is less than the lower limit. Then, in step S602, it is judged that the detection result of "the playing time of two consecutive pages p(n) and p(n+1) are both less than the lower limit value" is "yes" or "no".

If the judgment result of step S602 is "Yes", it means that it is detected that the respective play times of two consecutive pages p(n) and p(n+1) are less than the lower limit (that is, two consecutive pages p(n) , the playback speed of p(n+1) is increased), then it is judged that the arrangement reasoning model 200 needs to be optimized (ie, retrained), and the control signal S3 can be sent to activate the model optimization unit 400 .

Then, in step S603, the state controller can locate the previous page p(n-1) of the two consecutive pages p(n) and p(n+1) whose playback time is less than the lower limit, and the model optimization unit 400 The optimization will be adjusted for the previous page p(n-1).

Then, execute the observation agent mechanism in step S604, including: obtain the background attribute data D3 of the participants A1-A60 in the data set DS, and adjust the weight value of the previous page p(n-1) (for example, set its weight value to reduce).

Then, in step S605 , the model optimization unit 400 adjusts and optimizes (ie, retrains) the orchestration reasoning model 200 through reinforcement learning. For example, the content of the page p(n-1) can be adjusted with reference to the background attribute data D3 of the conference participants A1-A60 and/or the page p(n-1) can be adjusted to a later page sequence. Then, go back to step S601 to execute the state controller to determine whether the orchestration reasoning model 200 needs to be optimized.

On the other hand, if the judgment result of the state controller is "No" in step S602, which means that the orchestration reasoning model 200 does not need to be optimized, then perform step S604: maintain the current orchestration reasoning model 200 in the observation agent mechanism.

7A and 7B are flowcharts of a data processing method according to an embodiment of the present disclosure. The data processing method can be implemented based on each device or unit of the data processing system 1000 shown in FIG. 1 to FIG. 6 . Please refer to Figure 7A first. First, in step S701, according to the category, style and style of the initial presentation file B0, the classification label data D1 of the initial presentation file B0 is generated, and the playing time of each page of the presentation file B0 is analyzed to generate the playing time data of the initial presentation file B0 D2.

Then, in step S702, the meeting notice N0 is received, and the background and experience of the meeting participants A1-A60 are analyzed according to the information recorded in the meeting notice N0, so as to generate the background attribute data D3 of the meeting participants A1-A60.

Then, in step S703, the conference history data H0 is received from the database 2000, and the number of speech words, body movements and/or expressions of the participants A1-A60 in the conferences are analyzed and counted from the conference history data H0. To generate the participation degree data D4 of the conference participants A1-A60. In addition, the above-mentioned category marking data D1, playing time data D2, background attribute data D3, and participation level data D4 are collected to form a data set DS.

Then, in step S704 , preliminary training is performed on the orchestration reasoning model 200 according to the data set DS, so as to obtain the choreography reasoning model 200 after the preliminary training.

Please refer to Fig. 7B next. In step S705 , the choreography inference operation 250 is performed on the initial briefing file B0 with the choreography inference model 200 that has been preliminarily trained. For example, the pages that need to be corrected in the initial presentation file B0 can be listed, and the order of pages in the initial presentation file B0 can be adjusted to obtain the edited presentation file B1.

Then, in step S706, the presentation file B1 after editing is actually played during the meeting, and the voice Vocs of the participants A1~A60 are actually recorded during the meeting, and the above voice Vocs are converted into text and the number of words is counted , so as to count the number of words spoken by conference participants A1~A60 during the conference (for example: the number of speech words A1-1 of conference participant A1). And, the playback time of the edited briefing file B1 is actually recorded during the meeting.

Then, in step S707, according to the number of words spoken by the participants A1-A60 during the meeting and the playback time of the edited briefing file B1, it is judged whether to optimize (that is, retrain) the arrangement reasoning model 200 . Among them, it can analyze whether more than half of the conference participants A1~A60 have speech words of "zero", and detect whether there are two consecutive pages in the briefing file B1 whose playback time is less than the lower limit (for example, less than five seconds) ). If the judgment result of step S707 is "Yes", proceed to step S708: after the meeting ends, retrain the orchestration reasoning model 200 for optimization.

In step S708, an example of optimization by the arrangement reasoning model 200 is: it can locate the front page of two consecutive pages p(n) and p(n+1) whose playback time is less than the lower limit (that is, the playback speed is increased). A page p(n-1), and mark the number of pages p(n-1), p(n), p(n+1). And, reduce the weight value of the previous page p(n-1), so as to adjust the previous page p(n-1) to a later page sequence. In other words, it can be detected that the participants A1~A60 are not interested in the previous page p(n-1) and want to urge the next two pages p(n) and p(n+1) to play faster; therefore, for the previous page p(n-1) The content and page sequence of the page p(n−1) are adjusted, so as to optimize the layout reasoning model 200 .

On the other hand, if the determination result of step S707 is “No”, the current orchestration reasoning model 200 is maintained without optimization.

To sum up, the data processing system 1000 and the corresponding data processing method disclosed herein can adjust and arrange the initial presentation file B0 to obtain an edited presentation file B1 that is more in line with the interests of the conference participants A1-A60. Before the meeting, the data processing system 1000 can obtain the data set according to the type and style of the briefing file, the playing time of the briefing file, the background experience of the meeting participants, and the degree of participation of the meeting participants (number of speeches, body movements or expressions) DS: Preliminarily train the arrangement reasoning model 200 according to the data set DS, and arrange and adjust the initial briefing file B0 according to the arrangement reasoning model 200 after the preliminary training to obtain the arranged briefing file B1. When the meeting is actually going on, the meeting recording unit 300 can actually analyze the playback time of the briefing file B1 after the arrangement and the actual participation degree (such as the number of speeches) of the participants A1~A60, so as to optimize the arrangement reasoning model 200 (retraining). The disclosed data processing system 1000 and the corresponding data processing method can intelligently and dynamically adjust the arrangement reasoning model 200, so that the arranged briefing file B1 has the best content most suitable for the conference participants A1-A60.

Although the present invention has been disclosed above in detail with preferred embodiments and examples, it should be understood that these examples are meant to be illustrative rather than limiting. It is expected that those skilled in the art can conceive various modifications and combinations, and the various modifications and combinations fall within the spirit of the present invention and the scope of the appended patent application.

1000: data processing system 2000: database 100: data collection processing unit 200: layout reasoning model 250: layout reasoning operation 300: meeting record unit 310: briefing page number recording device 320: radio device 321: speech-text conversion device 330: Briefing playback time recording device 340: image capture device 400: model optimization unit B0: initial briefing file B1: edited briefing file N0: meeting notice N01: background, department and title N02: project plan and work experience N03: key Performance indicators N04: Self-input information A1~A60: Participants A1-1: Number of speech words A1-2: Body movements A1-3: Expressions H0, H1: Meeting history data D1: Classification and labeling data D2: Playing time data D3: background attribute data D4: participation level data DS: data set Voc: voice Img: image S2, S3: control signal F11, F12, F13, F14, F15, F16: feature p1, p2, p15, p20, p21, p22 , p30, p(n-1), p(n), p(n+1): pages S251~S253: steps S601~S605: steps S701~S708: steps

FIG. 1 is a block diagram of a data processing system according to an embodiment of the present disclosure. Figures 2A-2F are schematic diagrams of the data collection processing unit classifying, naming or labeling the initial briefing files. FIG. 2G is a schematic diagram of analyzing the time attribute of the initial briefing file by the data collection processing unit. FIG. 3 is a schematic diagram of analyzing the meeting participants by the data collection processing unit. FIG. 4 is a schematic diagram of the operation of an orchestration reasoning model according to an embodiment of the present disclosure. FIG. 5A is a block diagram of a meeting minutes unit according to an embodiment of the present disclosure. FIG. 5B is a schematic diagram of the operation of the sound receiving device and the image capturing device according to an embodiment of the present disclosure. FIG. 5C is a schematic diagram of the coordinated operation of the presentation playback time recording device and the presentation page number recording device according to an embodiment of the present disclosure. FIG. 6 is a flowchart of the operation of the model optimization unit according to an embodiment of the present disclosure. 7A and 7B are flowcharts of a data processing method according to an embodiment of the present disclosure.

1000: data processing system

2000: Database

100: data collection processing unit

200: Orchestrating Inference Models

300: Meeting Minutes Unit

400: Model optimization unit

B0: Initial briefing file

B1: Arranged briefing file

N0: meeting notice

H0, H1: Conference history data

D1: Classification and labeling information

D2: Playing time data

D3: Background attribute data

D4: Participation level data

DS: data set

Voc: Voice

Img: Image

S2, S3: control signal

Claims (20)

A data processing system comprising: an orchestration inference model for receiving an initial briefing file and a data set, and performing an orchestration inference operation based on the initial briefing file and the data set to generate an edited briefing file; and A meeting record unit, used to record the playback time of the briefing file after the arrangement, and record the voices of multiple participants in the meeting and count the number of words spoken by these participants, and according to the playback time of the briefing file after the arrangement and the number of words spoken by those participants to determine whether to retrain the arrangement reasoning model, Wherein, the data set is at least related to the type and style of the initial briefing file and the background and experience of the participants, and the data set is used to initially train the arrangement reasoning model. The data processing system as claimed in claim 1, wherein the arrangement reasoning operation performed by the arrangement reasoning model includes: listing the pages in the initial presentation file that need to be corrected, and adjusting the order of pages in the initial report file. The data processing system as described in Claim 1, further comprising: A data collection processing unit, used to generate a classification label data and a play time data of the initial briefing file; Wherein, the category marking data is related to the type and style of the initial briefing file, and the data set at least includes the category marking data and the playing time data. The data processing system as described in Claim 3, wherein the data collection processing unit is further used for: receiving a meeting notice, and generating a background attribute data of the meeting participants according to the meeting notice; Wherein, the background attribute information is related to the background and experiences of the conference participants, and the data set further includes the background attribute information. The data processing system as described in Claim 4, wherein the data collection processing unit is further used for: Obtain a meeting history data from a database, and generate a participation level data of those participants in the meeting according to the meeting notice and the meeting history data; Wherein, the participation level data is related to speech words, body movements and/or expressions of the participants, and the data set further includes the participation level data. The data processing system as described in Claim 1, further comprising: a model optimization unit for retraining the orchestration reasoning model; Wherein, when the playback time of the programmed briefing file is shortened and/or more than half of the conference participants have "zero" speech words, the model optimization unit is activated to retrain the programming reasoning model. The data processing system as described in claim 6, wherein when the playback time of two consecutive pages of the presentation file after the editing is less than a lower limit, it is judged that the playback time of the presentation file after the editing is shortened, and the model optimization unit is started to retrain the orchestrated inference model. The data processing system as described in Claim 7, wherein the model optimization unit includes: a state controller, used to locate the previous page of the two consecutive pages whose playback time is less than the lower limit; and An observation agent mechanism is used to adjust the weight value of the previous page, so as to adjust the previous page to a lower page sequence. The data processing system as described in Claim 8, wherein the meeting record unit includes: a radio device, used to record the voices of the conference participants, and convert the voices of the conference participants into text to count the number of words spoken by the conference participants; and A presentation playback time recording device is used to record the playback time of the edited briefing file, and detect whether the respective playback times of the two consecutive pages are less than the lower limit. The data processing system as described in Claim 8, wherein the meeting record unit further includes: A briefing page number recording device is used to record the number of pages of each page of the edited briefing file, and mark the number of pages of the two consecutive pages and the previous page whose playing time is less than the lower limit. A data processing method, comprising the following steps: receiving an initial presentation file and a data set, wherein the data set is at least related to the type and style of the initial presentation file; initially training an orchestration inference model based on the data set; performing a choreography inference operation by the choreography inference model based on the initial profile and the data set to generate a choreography profile; Record the playing time of the briefing file after the arrangement; recording the voices of a plurality of conference participants and counting the number of words spoken by these conference participants, wherein the collection of data is at least related to the background and experience of these conference participants; and It is judged whether to retrain the arrangement reasoning model according to the playback time of the arranged briefing file and the speech words of the participants. The data processing method as described in Claim 11, wherein the orchestration and inference operation includes the following steps: List the pages in the initial briefing document that require revision; and Adjust the page order of this initial presentation file. For the data processing method described in Claim 11, before the step of receiving the initial briefing file and the data collection, the data processing method further includes the following steps: generating a category labeling data and a playing time data of the initial presentation file, wherein the category labeling data is related to the type and style of the initial presentation file; and At least the classification marking data and the playing time data form the data set. According to the data processing method described in Claim 13, before or after the step of receiving the initial briefing file and the data collection, the data processing method further includes the following steps: receiving a meeting notice, and generating a background attribute data of the meeting participants according to the meeting notice, wherein the background attribute data is related to the background and experience of the meeting participants; and Add the background attribute data to the data collection. According to the data processing method described in Claim 14, before or after the step of receiving the initial briefing file and the data collection, the data processing method further includes the following steps: Obtain a meeting history data, and generate a participation level data of those participants according to the meeting notice and the meeting history data, wherein the participation level data is related to the number of speech words, body movements and/or expressions; and Add the engagement level data to the data collection. The data processing method as described in Claim 11, wherein the step of judging whether to retrain the orchestration reasoning model includes: When the playback time of the edited briefing file is shortened and/or more than half of the conference participants have "zero" speeches, it is determined that the edit reasoning model needs to be retrained. The data processing method as described in claim 16, wherein when the playback time of two consecutive pages of the edited briefing file is less than a lower limit, it is judged that the playback time of the edited briefing file is shortened and the arrangement is retrained reasoning model. The data processing method as described in Claim 17, wherein the step of retraining the orchestration reasoning model comprises: Locating the previous page of the two consecutive pages whose playing time is less than the lower limit; and The weight value of the previous page is adjusted to adjust the previous page to a lower page order. The data processing method as described in claim item 18, wherein the step of counting the number of words spoken by those participants in the conference includes: transcribe the speech of those meeting participants into text; and Count the number of words in these texts. The data processing method described in Claim 18 further includes the following steps: record the number of pages of each page of the formatted presentation file; and The number of pages of the two consecutive pages and the previous page whose playing time is less than the lower limit value is marked.

Images