Patent Application
20200051450
The invention relates to an audio-visual teaching platform, particularly to achieve the recommendation and adaptability evaluation of learning information by collecting interactive information of users, and to analyze the interactive audio-visual teaching platform between users and audio-visual knowledge content.
With the development of network and personal electronic equipment, the old teaching methods such as after-class school or tutor have been gradually replaced by multimedia teaching. Among them, in the field of individual learning or enterprise education and training, online multimedia teaching is not limited to fixed teaching time and fixed teaching location, can stop at any time, replay, fast forward and other characteristics are also the main reasons for its popularity.
Current multimedia teaching often transmits learning information of course-related audio-visual knowledge content from the teaching system end to the user end. However, it is hard to measure or get feedback on the real participation and learning effectiveness of the users, nor to adjust learning content properly by the learning levels of the users, resulting in one-way learning information transmission.
Further when the enterprise uploads the audio-visual knowledge content independently by the internal staff and chooses the appropriate audio-visual knowledge content therein to make a course for education and training. This practice will produce excessive fragmentary audio-visual knowledge content, which not only needs a lot of manpower to organize, but also is difficult to link with the audio-visual knowledge content of formal and systematic education and training, resulting in the bewildered dilemma of trainees in enterprises.
Although there are some data such as the number of times and hours of playing learning information as a measure of learning effectiveness, it is still unable to get feedback from the users and further adjust the learning content appropriately. Take the education and training of enterprises as an example, the users usually use idle computers to broadcast learning information to lengthen the watching hours, but they do not really participate in learning or make a response to learning behavior.
In addition, the learning effectiveness and measurement of learning information have different meanings of user interaction in different fields, types and even different lecturers. Therefore, there must be a weighted measurement of the evaluation method.
Thus, how to make use of historical learning records and collect and analyze feedback information of the users in the learning process to integrate various sources of the audio-visual knowledge content, including formal education and training content, external digital platform knowledge content, and internal users upload knowledge content for planning the optimum curriculum structure, and to further carry on recommending and assessing the adaptability of the learning information based on this framework has become the focus that all walks of life want to improve urgently.
In view of the problems described in the previous techniques, one of the purposes of the present invention is to provide an audio-visual teaching platform by which an analysis subsystem can provide a learning behavior data so that the recommendation subsystem can use the learning behavior data to generate at least one presumptive learning mode, thereby generating a recommended learning combination. In this way, when the training subsystem receives the recommended learning combination, it will set up the audio-visual knowledge content according to the recommended learning combination to transmit the set audio-visual knowledge content to a database for storage and transmission to a user for watching and learning.
Accordingly, the related audio-visual knowledge content is recommended to the users of the same or similar presumptive learning mode, and the interactive users between the users and the audio-visual knowledge content are analyzed, as well as the time points at which the concentration of the users in the learning process decreases and improves, so that the users can really watch the audio-visual knowledge content and enhance their interest in learning.
In view of the problems described in the previous technology, another purpose of the present invention is to use the historical learning records of the users and collect and analyze feedback information in the learning process to achieve the adaptability evaluation of learning information.
Another purpose of the present invention is to provide an analysis method of an audio-visual teaching platform, which mainly includes user-defined interactive element, user analysis module, audio-visual content analysis module, message analysis module, time-sequential interactive module and integrated multi-sequence analysis module.
The user-defined interactive element is set on an audio-visual knowledge content interface and contains interactive factors, such as asking questions, answering questions, taking notes, marking key points, emoticons, fast forward and rewind, each of the interactive factors provides the input of interactive data.
The user analysis module uses plural historical learning behavior data of all users in a database to calculate a distance and clustering according to each of the plural historical learning behavior data, and categorizes the current users into their corresponding user clusters to generate the user clustering data related to clustering results.
The object in image data belonging to the audio-visual knowledge content is marked through an audio-visual content analysis module to generate the image object sequence related to each of time points in the audio-visual knowledge content, carries on an audio frequency analysis of the audio data belonging to the audio-visual knowledge content, calculates a pitch of each of audio frames, and generates the audio sequence related to each of the time points in the audio-visual knowledge content.
The message analysis module uses texts extracted from the database to carry on purpose classification for interactive data to generate the message sequence related to the purpose of the interactive data.
The time-sequential interactive module correlates the input time of the interactive data and the corresponding time points in the time axis of the audio-visual knowledge content and generates the interactive time points, and then combines each of the interactive time points in the time axis of the audio-visual knowledge content to generate the time-sequential interactive sequence.
The integrated multi-sequence analysis module generates a learning behavior data of a current user related to an interactive factor at any one of the interactive time points on the time axis of the audio-visual knowledge content by using a user clustering data, an image object sequence, an audio sequence, a message sequence and a time-sequential interactive sequence.
The user-defined interactive element further sets a weight of the interactive factors, the learning behavior data corresponding to each of user clusters and the interactive factors are generated through a correlation between the integrated multi-sequence analysis module and the user clustering data.
Another purpose of the present invention is to further provide an analysis method of an audio-visual teaching platform which comprises the steps of:
inputting an interactive data into an interactive factor on a user-defined interactive element set on an audio-visual knowledge content interface;
using user analysis module to calculate a distance and clustering of plural historical learning behavior data of all users in a database according to each of the plural historical learning behavior data, and categorizing current users into corresponding user clusters to generate a user clustering data;
marking object in an image data belonging to an audio-visual knowledge content by an audio-visual content analysis module to generate an image object sequence related to each of time points in the audio-visual knowledge content;
calculating a pitch of each of audio frames of audio data that belongs to the audio-visual knowledge content by the audio-visual content analysis module for audio frequency analysis to generate an audio sequence related to each of the time points in the audio-visual knowledge content;
extracting texts in the database, and carrying on a purpose classification of the interactive data by a message analysis module to generate a message sequence related to a purpose of the interactive data;
taking an input time of the interactive data with each of the corresponding time points in the audio-visual knowledge content to generate each of the interactive time points by a time-sequential interactive module, and combining each of the interactive time points in a time axis of the audio-visual knowledge content to generate a time-sequential interactive sequence; and
correlating the user clustering data, the image object sequence, the audio sequence, the message sequence and the time-sequential interactive sequence with the interactive data by an integrated multi-sequence analysis module to generate a learning behavior data at any one of the time points in the audio-visual knowledge content.
The invention provides a time-sequential interactive module in the analysis method of the audio-visual teaching platform, which collects each of the interactive time points on the time axis of the audio-visual knowledge content and the corresponding learning behavior data to generate a long-short term memory.
The current users are compared with the long-short term memory, wherein the current users are categorized into user clusters with similar learning styles in the long-short term memory according to the learning behavior data.
The long-short term memory is used to judge the current users that belong to the user clusters, and evaluates the concentration indicator according to the distribution of each of the interactive time points on the time axis of the user clusters in the audio-visual knowledge content and the corresponding learning behavior data of each of the interactive time points.
The long-short term memory is used to judge the current users that belong to the user clusters, and activates the interactive factor on the user-defined interactive element according to a time interval of lower concentration indicator on the time axis of the user clusters in the audio-visual knowledge content to provide input of the interactive data.
In view of the problems described in the previous techniques, the present invention further aims at exploiting the subsequent correlated audio-visual knowledge content related to the knowledge content from multiple sources in the past through experiment and validation processes by using the behavior of at least one user who has watched the plural audio-visual knowledge content from multiple sources in the past, which makes it possible to achieve recommendation of learning data and application of adaptability without need too much manpower to organize the plural audio-visual knowledge content.
A further purpose of the present invention is to provide a recommendation subsystem of an audio-visual teaching platform, which includes an exploration module, an experiment module and a verification module. The exploration module generates at least one presumptive learning mode according to a learning behavior data from which at least one user watches the plural audio-visual knowledge content from multiple sources. The experiment module generates the subsequent learning experiment combination according to the presumptive learning mode, and the verification module validates whether the subsequent learning experiment combination meets validation requirements according to one of the learning modes. When it meets the requirements, the subsequent learning experiment combination is taken as a recommended learning combination.
The steps of using the exploration module to generate at least one presumptive learning mode according to a learning behavior data of at least one user watching the plural audio-visual knowledge content from multiple sources further include: categorizing each of the plural audio-visual knowledge content into at least one knowledge cluster according to its source and learning theme, categorizing the user to each audio-visual knowledge; taking at least one learning behavior of the plural audio-visual knowledge content as a key indicator; capturing the time-sequential learning behavior sequence data of each of the at least one user in the at least one knowledge cluster, wherein the data format is the user access and interactive record of knowledge content at that time point; and using decision tree classification model to distinguish the learning behavior of the at least one user in the at least one knowledge cluster and generating learning modes; finally, calculating the key indicator of each of the learning modes and taking the output as the at least one presumptive learning mode.
A further purpose of the present invention is to provide a recommendation method of an audio-visual teaching platform which includes the following steps of:
using the exploration module to generate at least one presumptive learning mode according to the learning behavior data of the at least one user watching the plural audio-visual knowledge content from multiple sources; and using the experiment module to generate the subsequent learning experiment combination according to the presumptive learning mode; and then using the verification module validates whether the subsequent learning experiment combination meets validation requirements according to one of the learning modes. When the verification module judges that the subsequent learning experiment combination meets the requirements, the verification module takes the subsequent learning experiment combination as the recommended learning combination, and recommends it to other users with the same presumptive learning mode, so that the plural audio-visual knowledge content from multiple sources can be optimized for cohesive learning.
The recommendation subsystem of the audio-visual teaching platform further includes an optimum practice module which converts the time-sequential learning behavior sequence data and the identity information of all users to a series of variables, in which the identity information of the user includes user name, gender, age or name of affiliated unit to which the user belongs. The time-sequential learning behavior sequence data can be the user message time, message content and marked object. The optimum practice module takes each user as a node and uses the series of variables to calculate clustering distances between each user and uses this as a clustering benchmark, and classifies each user who meets thresholds of different clustering benchmarks into at least one clustering group, or the user whose average distance of one of the clustering distances is the smallest in each of the at least one clustering group is further searched as a group opinion leader of this clustering group.
Moreover, the optimum practice module further uses frequent pattern mining to calculate non-clustered user who is most in line with a browsing recommended learning mode under each of the knowledge clusters, and actively through one of the recommendation, matching, competition mechanisms, to make the non-clustered user approach one of the clustering groups.
Refer to
The audio-visual teaching platform is mainly composed of analysis subsystem 100, recommendation subsystem 200 and training subsystem 300. The analysis subsystem 100 is used to analyze user clustering data, image object sequence, audio sequence, message sequence and time-sequential interactive sequence, etc. to generate learning behavior data 101 related to an interactive factor at any one of the interactive time points on a time axis of audio-visual knowledge content 301.
The recommendation subsystem 200 is connected with the analysis subsystem 100 to generate at least one presumptive learning mode according to the learning behavior data 101 of at least one user 400 watching plural audio-visual knowledge content 301 from multiple sources by the procedures of exploration, experiment and validation. Then a subsequent learning experiment combination is generated by the experiment procedures according to the presumptive learning mode. Finally, whether the subsequent learning experiment combination meets the validation requirement is verified. The subsequent learning experiment combination is taken as the recommended learning combination 201 when it meets the requirement.
The training subsystem 300 is connected with the recommendation subsystem 200 and the analysis subsystem 100. By receiving the recommended learning combination 201, each of the interactive time points within each of the plural audio-visual knowledge content 301 can be preset according to the recommended learning combination 201, and a knowledge map can be generated according to the recommended learning combination 201 to transmit each of the interactive time points and knowledge map within each of the plural audio-visual knowledge content 301 to the database 500. The training subsystem 300 provides at least one of the plural audio-visual knowledge content 301 which is preset to a user 400 to provide the at least one of the plural audio-visual knowledge content 301 to the user 400 for watching and learning.
In this way, the user 400 can carry on the actions of watching, learning and interaction through the plural audio-visual knowledge content 301 provided by the training subsystem 300. When the user 400 watches the plural audio-visual knowledge content 301, the analysis subsystem 100 and the recommendation subsystem 200 will also carry on the follow-up actions, and then learn and optimize the function of analysis and recommendation.
Refer to
The analysis subsystem of the audio-visual teaching platform includes a user analysis module 120, a message analysis module 130, an audio-visual content analysis module 140, a time-sequential interactive module 150 and an integrated multi-sequence analysis module 160. These modules can be edited by programming languages and stored in electronic devices or storage media devices connected by the electronic devices (for example, network file servers, disk drives or portable disks, etc.).
The user analysis module 120 uses plural historical learning behavior data of all users in the database and data clustering techniques, such as K-means and hierarchical clustering, to calculate distance and clustering according to the plural historical learning behavior data of current users, to categorize the current users into their corresponding user clusters, and the clustering results are taken as the user clustering data, wherein the user clustering data are related to the user past learning process or the use of platform-related statistical information items, such as: question answering rate, average watching duration, number of friends and so on.
Taking K-means as an example, assuming that the two variables of question answering rate and average watching duration of the user are respectively taken as the variables of clustering, the relationship can be expressed in a two-dimensional coordinate system. Firstly, k centers are randomly selected, and the distance between each data point xj and each of the centers μi (using Euclidean distance) is calculated. Clustering is based on the centers nearest to each data point. Then, a new center point S is recalculated according to the clustering results. The following formula is used to iterate the above process to get a minimum value J:
The message analysis module 130 uses the corresponding texts extracted from the database and carries on a purpose classification based on the interactive data input by the current user, such as using an intent classification method and generating a message sequence, wherein the intent classification is based on the texts information input by the user and further judges the purpose of entering the text message. When using a long-short term memory (LSTM) after defining several categories in the database, such as airplane, train, mass rapid transit (MRT) and other types of vehicles, the most likely type of vehicle to be taken by the user is judged according to the content of the text message input by the user. For example, if the user enters “I want to go from Taiwan to Japan”, LSTM will output “airplane”. The purpose of the intent analysis is to judge the content of the message sequence, such as questions, complaints and gossip from the user.
The audio-visual content analysis module 140 marks the object in the image data that belong to the input of the plural audio-visual knowledge content, such as identifying what object is in the image data, generates the image object sequence related to each of the time points in the plural audio-visual knowledge content, and carries on an audio frequency analysis of audio data belonging to the input of the plural audio-visual knowledge content. Firstly, the whole audio data is cut into several audio frames, and then the original audio is down sampling several times by a harmonic product spectrum method. The compressed audio of the original audio after sampling is merged to highlight the high point of the fundamental frequency, calculate a pitch of each of the audio frames, eliminate unstable pitch and smoothing, and generate audio sequence related to each of the time points in the plural audio-visual knowledge content such as pitch and frequency at each of the time points.
The time-sequential interactive module 150 takes the time point when inputting the interactive data and the corresponding time point in the time axis of the plural audio-visual knowledge content when inputting the interactive data as an interactive time point, and generates a time-sequential interactive sequence that contains interactive time points.
The integrated multi-sequence analysis module 160 combines the above user clustering data, image object sequence, audio sequence, message sequence and time-sequential interactive sequence to produce a learning behavior data. From the learning behavior data, we can know the interactive results generated among the current users and the mentioned users in the user clusters and the images or sounds in the plural audio-visual knowledge content at any one of the interactive time points in the time axis of the plural audio-visual knowledge content. A long-short term memory can also be built from the distribution of the interactive time points in the time axis of the plural audio-visual knowledge content.
Refer to
S101: playing an audio-visual knowledge content from user, wherein a user-defined interactive element sets on an audio-visual knowledge content interface;
S102: inputting an interactive data into an interactive factor in the user-defined interactive element from the user;
S103 using user analysis module to calculate a distance and clustering of plural historical learning behavior data of all users in a database according to each of the plural historical learning behavior data, and categorizing current users into corresponding user clusters and taking this clustering result as user clustering data;
S104: marking object in an image data belonging to an audio-visual knowledge content by an audio-visual content analysis module to generate an image object sequence related to each of time points in the audio-visual knowledge content;
S105: calculating a pitch of each of audio frames of audio data that belongs to the audio-visual knowledge content by the audio-visual content analysis module for audio frequency analysis to generate an audio sequence related to each of the time points in the audio-visual knowledge content;
S106: extracting texts in the database, and carrying on a purpose classification of the interactive data by a message analysis module to generate a message sequence related to a purpose of the interactive data;
S107: taking an input time point of the interactive data with a corresponding time point on a time axis of the audio-visual knowledge content as an interactive point and collecting the interactive time points on the time axis to generate a time-sequential interactive sequence; and
S108: combining the user clustering data, the image object sequence, the audio sequence, the message sequence and the time-sequential interactive sequence by an integrated multi-sequence analysis module to generate learning behavior data corresponding to the interactive time points in the time axis of the audio-visual knowledge content.
When the user re-starts the interactive factors in the user-defined interactive element, the above steps of S102-S108 are repeated to generate the corresponding learning behavior data at the interactive time points in the time axis of the audio-visual knowledge content.
The above steps of S102-S107 are performed after receiving the interactive data input from the current user in the interactive factors, and the execution order is not subject to the limits.
Refer to
The schematic diagram for establishing the long-short term memory:
The time-sequential interactive module collects the interactive time points on the time axis of the audio-visual knowledge content and the corresponding learning behavior data and builds a long-short term memory. In other words, user clustering data, image object sequence, audio sequence, message sequence and time-sequential interactive sequence are used as input parameters to establish a long-short term memory that can achieve self-learning, concentration assessment and adaptability.
Self-learning of the long-short term memory:
The long-short term memory compares the learning behavior data input by the current users with the long-short term memory. According to the learning behavior data, the current users are categorized into user clusters with similar learning styles in the long-short term memory for self-learning.
This long-short term memory can be learned with the increase of the amount of learning behavior data of users in the database, and the current users can be immediately categorized into corresponding user clusters to carry on subsequent predictions.
The concentration assessment of the long-short term memory:
The long-short term memory is used to judge the current users that belong to the user clusters, and evaluates the concentration indicator according to the distribution of each of the interactive time points on the time axis of the user clusters in the audio-visual knowledge content and the corresponding learning behavior data of each of the interactive time points.
By clustering the current users or the user clusters to which the current users belong, the concentration indicator changes with the corresponding time axis in the audio-visual knowledge content that provides a basis for further predicting the possible interactive behavior generated or adjusting the audio-visual knowledge content for the system administrator.
Concentration prediction and incentive mechanism of the long-short term memory:
The long-short term memory is used to judge the current users that belong to the user clusters, and activates the interactive factors on the user-defined interactive element according to a time interval of lower concentration indicator on the time axis of the user clusters in the audio-visual knowledge content to provide input of the interactive data. By increasing the weight of the interactive factors increases the motivation of users to input interactive data and promote learning behavior data.
By using the long-short term memory to evaluate the user concentration instead of using watch hours as the basis to evaluate the learning effectiveness can predict the time points at which the concentration indicator of current users may decline in the audio-visual knowledge content, and activate the interactive factors at that time points to provide incentives such as rewards, points, etc. Getting interactive feedback and increasing attention from current users can effectively improve the problem of playing learning information on idle computers, so as to enable current users to really watch the content of the audio-visual knowledge content and enhance their interest.
Refer to
The exploration module 210 receives various learning behavior data from different users watching plural audio-visual knowledge content from different sources among the same or different users, and generates at least one presumptive learning mode according to at least one of these learning behavior data. The users can be another electronic device, such as personal computers, laptops, tablets, smart phones, etc. with connecting networks (e.g. local area network or Internet).
The exploration module 210 categorizes each of the plural audio-visual knowledge content into at least one knowledge cluster according to its source and learning theme. Furthermore, the exploration module 210 can use existing exploration techniques based on semantic or literary meanings to analyze the text or voice content of each of the plural audio-visual knowledge content, and automatically categorize each of the plural audio-visual knowledge content into different knowledge clusters, or receive the categorization information of the system administrator or user, and categorize each of the plural audio-visual knowledge content into at least one knowledge cluster according to its source and learning theme.
For example, junior high school and elementary school students have knowledge clusters of different disciplines such as Chinese, English, mathematics, etc. Enterprises have knowledge clusters of different education and training disciplines, such as newcomer training, corporate culture, etc. For example, the knowledge cluster named “Insurance” may include the plural audio-visual knowledge content such as long-term care insurance films uploaded by users themselves, accidental insurance films loaded by system administrators in the enterprise curriculum library, fire affairs films uploaded by users from third-party audio-visual websites, car accidents films, etc. Or, the plural audio-visual knowledge content from different sources, such as gender equality, work rights and vacation regulations, can be categorized together as a knowledge cluster named “Labor Rights and Interests”.
Alternatively, the plural audio-visual knowledge content such as English vocabulary, English grammar, English sentence patterns, Japanese vocabulary, Japanese grammar and Japanese sentence patterns can be categorized together as a knowledge cluster named “Foreign Languages Learning”. Otherwise, the plural audio-visual knowledge content such as English vocabulary, English grammar and English sentence patterns can be categorized as a knowledge cluster named “English Learning”, or Japanese vocabulary, Japanese grammar, and Japanese sentence patterns can be categorized as a knowledge cluster named “Japanese Learning”.
From the above, it can be seen that the sources of the plural audio-visual knowledge content can be at least made up of user-made teaching films, teaching films uploaded by system administrators, or films downloaded or linked by audio-visual websites. All knowledge clusters are stored in the database.
Further, the exploration module 210 takes at least one learning behavior or a self-defined operational formula of the user for each of the plural audio-visual knowledge content as a key indicator. The learning behavior can be the test scores after learning, the number of interaction between users and instructors, the number of questions asked, etc. The self-defined operational formula can be described as a combination of any of the above two learning behaviors, or a combination of any of the above two learning behaviors calculated with different weights. For example, the key indicator is that the test scores after learning are above 90 points; or the test scores after learning are above 80 points, and the number of interaction between users and lecturers is more than three times; or the key indicator is that the test scores after learning are above 80 points accounted for 30% of the weight, and the number of interaction between users and lecturers is more than three times accounted for 30% of the weight, and the number of questions asked is more than four times accounted for 40% of the weight.
Furthermore, the exploration module 210 captures the time-sequential learning behavior sequence data of each user in the knowledge cluster, and uses decision tree classification model to distinguish the learning behavior of each user in each knowledge cluster according to each of the key indicators, thus generating learning modes. Finally, calculate an optimum of the key indicators or a key threshold set for each of the learning modes, taking output as at least one presumptive learning mode.
Moreover, the learning behavior refers to an access behavior at each of the time points and an interactive data for each of the plural audio-visual knowledge content when each user watching the plural audio-visual knowledge content from different sources, wherein the access behavior includes normal playback, backward playback, various times of fast playback, pause, etc., while the interactive behavior includes commenting, asking questions, answering questions, taking notes, marking the key points, etc.
The experiment module 220 selects one of the learning modes and one of the knowledge clusters as the experimental learning mode, from which a subsequent learning experiment combination is generated. The subsequent learning experiment combination includes at least one of the audio-visual knowledge content, the learning behavior and a key experimental indicator. In addition, the audio-visual knowledge content of the subsequent learning experiment combination can be the one that the user has watched or has not watched, and the learning behavior can be the same or different from a selected learning mode. The key experimental indicator can be the same or different from the key indicator of the selected learning mode.
Furthermore, the experiment module 220 can open connection interfaces to allow external testing groups to make use of the subsequent learning experiment combination to carry on the experiment. The external testing group can enter the experiment module 220 for the users of a community website to learn. The experiment module 220 judges whether the users of the testing group in the process of experiment have browsed the audio-visual knowledge content and practiced their learning behavior, and whether they have met the requirement of the key experimental indicator. The experiment is successful if the users browse the audio-visual knowledge content of the subsequent learning experiment combination, practice the learning behavior and meet the requirement of the key experimental indicator. Otherwise, the experiment module 220 needs to select the other learning modes of the selected knowledge cluster and repeat the experiment according to the above process until it succeeds.
The verification module 230 selects at least one learning mode to validate whether the subsequent learning experiment combination meets validation requirements. When the verification module 230 validates that the subsequent learning experiment combination meets the requirements, the subsequent learning experiment combination is used as the recommended learning combination. The validation method of the verification module 230 for the subsequent learning experiment combination is to define at least one learning mode accepted by the verification module 230 as prior distribution data, and the learning mode selected by the verification module 230 must be different from the learning mode selected by the experiment module 220, and using this to calculate posterior distribution data. After completing the experiment in the experiment module 220, the distribution distance between the posterior distribution data after random sampling and the time-sequential learning behavior sequence data of the subsequent learning experiment combination is calculated.
The aforementioned distribution distance algorithm is different from the node distance mentioned above. The so-called distribution distance refers to the time-sequential learning behavior sequence data of the actual learning behavior of the subsequent learning experiment combination. When it is similar to the time-sequential learning behavior sequence data of the posterior distribution data, it indicates that the distribution distance is close, otherwise the distance is far. The most commonly used method is KL divergence (Kullback-Leibler Divergence), and the distribution distance is defined as
From the two probability distributions p(x) and q(x) in the above equation formula can be known that when the difference between these two distributions is large, p(x)/q(x) is quite large, and the difference between these two distributions is small, p(x)/q(x) is small. It is defined as whether the similarity between these two distributions meets a validation threshold or not. If it is judged that it meets the validation threshold, the subsequent learning experiment combination is taken as the recommended learning combination.
If it is judged that it does not meet the validation threshold, it is still considered as a failure of the experiment. It is necessary to regenerate a subsequent learning experiment combination, and carry on the experiment and validation once again. Regardless of the validation is successful or not, the invention can reenter the exploration module 210 and the experiment module 220 at any time to recalculate the learning mode, the presumptive learning mode and the subsequent learning experiment combination, and carry on the experiment and validation.
In the present invention, the optimum practice module converts the time-sequential learning behavior sequence data and identity information of users to a series of variables. The optimum practice module takes each of the users as a node to calculate clustering distances between each of the series of variables and uses this as a clustering benchmark, and classifies each of the users who meets the thresholds of different clustering benchmarks into at least one clustering group, or the user whose average distance of one of the clustering distances is the smallest in each of the at least one clustering group is further searched as a group opinion leader of this clustering group.
The user identity information includes the user name, gender, age or name of affiliated unit to which the user belongs, etc. The time-sequential learning behavior sequence data can be the message time, message content and marked object, etc. of the user. The learning behavior of the user can be the operations on the playing way of films, such as pause, rewind, message, and each user has different operations for different types of films. To illustrate how the optimum practice module converts the time-sequential learning behavior sequence data and identity information of the user to a series of variables, the following is illustrated how an employee watches films of knowledge clusters that are categorized as newcomer training:
The employee presses pause at 1:05 of the film playing time without leaving a message. The affiliated unit of the employee is the management department, and the series of variables can be defined as follows:
x1=65, x2=pause, x3=empty, x4=newcomer training, x5=management department; wherein
x1=65, means that there is a learning behavior when the film is played to 1 minute and 5 seconds; x2=pause, means that learning behavior pauses to play the film; x3=empty, means that there is no message content at this time. If there is any message, message content will be filled in.; x4=newcomer training, means that the employee is watching films of knowledge clusters that are categorized as newcomer training; x5=management department, means that the employee is listed in the affiliated unit of the enterprise.
If the employee leaves a message when the film is played to 2 minutes and 12 seconds, the series variables can be defined as follows:
x1=132, x2=message, x3=message content, x4=newcomer training, x5=management department; wherein
x1=132, means that there is a learning behavior when the film is played to 2 minutes and 12 seconds; x2=message, means that learning behavior is a message, x3=message content, means that the message at this time is the message content; x4=newcomer training, means that the employee is watching films of knowledge clusters that are categorized as newcomer training; x5=management department, means that the employee is listed in the affiliated unit of the enterprise.
The above examples are just for illustration, which are not limited in practical application. For example, x4=newcomer training, in addition to representing the knowledge cluster to which the film belongs, a series of variables of the chapters and types of the film can also be added, as well as a series of variables of the ratings after watching by the trainees, to help the optimum practice module to establish the accuracy of clustering.
Moreover, the optimum practice module further uses frequent pattern mining to calculate non-clustered user who is most in line with a browsing recommended learning mode under each of the knowledge clusters, and actively through one of the recommendation, matching, competition mechanisms or any combination of the two or more to make the non-clustered user approach one of the clustering groups.
To sum up, please refer to
Moreover, after selecting the recommended learning mode under each knowledge cluster, if it is to be recommended to a certain group, the group opinion leader will be recommended first, and then let group opinion leader recommend it to other users of the clustering group they belong to, so that users of the same clustering group can have the opportunity to achieve the learning effect as the recommended learning mode. Or it can be directly recommended to the non-clustered users.
Refer to
S21: using an exploration module to generate at least one presumptive learning mode according to a learning behavior data of at least one user watching each of plural audio-visual knowledge content from multiple sources; the following steps are also included in this step:
S211: defining knowledge cluster: the exploration module that categorizes each of the plural audio-visual knowledge content into at least one knowledge cluster according to each of multiple sources and a same or similar learning theme; the exploration module can analyze the text or voice content of each of the plural audio-visual knowledge content using existing exploration techniques based on semantic or literary meanings, and automatically categorize each of the plural audio-visual knowledge content into different knowledge clusters, or receive categorization information of the system administrator or user, and categorize each of the plural audio-visual knowledge content into the at least one knowledge cluster according to its source and learning theme.
S212: defining key indicator: the exploration module that takes at least one learning behavior or a self-defined operational formula for each of the plural audio-visual knowledge content as a key indicator; the learning behavior can be the test scores after learning, the number of interaction between users and instructors, the number of questions asked, etc., the self-defined operational formula can be described as a combination of any of the above two learning behaviors, or any combination of the above two learning behaviors under different weights.
S213: generating at least one presumptive learning mode from the exploration module: the exploration module that captures each time-sequential learning behavior sequence data of each user in the knowledge cluster, and using decision tree classification model to distinguish the learning behavior of each user in each knowledge cluster according to each key indicator, thus generating learning modes, and then calculating an optimum of the key indicator or a key threshold set for each of the learning modes, and taking output as the at least one presumptive learning mode.
S22: generating a subsequent learning experiment combination from the experiment module according to the presumptive learning mode; the following steps are also included in this step:
S221: using the experiment module to select one of the learning modes as an experimental learning mode and one of the knowledge clusters, from which the subsequent learning experiment combination is generated; the subsequent learning experiment combination includes at least one of the audio-visual knowledge content, the learning behavior and a key experimental indicator, in addition, the audio-visual knowledge content of the subsequent learning experiment combination can be the one that the user has watched or has not watched, and the learning behavior can be the same or different from a selected learning mode, the key experimental indicator can be the same or different from the key indicator of the selected learning mode.
S222: using the experiment module to judge whether other users have browsed the audio-visual knowledge content of the subsequent learning experiment combination and practiced their learning behavior, if they carry on step S223, otherwise they carry on step S221; other users can open connection interfaces to allow external testing groups to make use of the subsequent learning experiment combination to carry on experiment.
S223: using the experiment module to judge whether the subsequent learning experiment combination meets the requirements of the key experimental indicator, if so, the experiment being successful, otherwise carrying on step S213.
Since it is not clear whether the experiment of the subsequent learning experiment combination which is successful through the experiment module can meet the learning purpose of each knowledge cluster, in order to make the subsequent learning experiment combination more close to the learning purpose of each knowledge cluster, the following steps are carried on after the above step S22:
S23: using the verification module to verify whether the subsequent learning experiment combination meets the validation requirements according to one of the presumptive learning modes, and the learning mode selected by the verification module that must be different from the learning mode selected by the experiment module, when the verification module judges that the subsequent learning experiment combination meets the requirements, the verification module taking the subsequent learning experiment combination as the recommended learning combination, otherwise, the experiment being deemed to have failed and carrying on step S213; the validation method of the verification module for the subsequent learning experiment combination including the following steps of:
S231: defining at least one of the learning modes accepted by the verification module as a prior distribution data and using this to calculate a posterior distribution data;
S232: calculating a distribution distance between the posterior distribution data after random sampling and the time-sequential learning behavior sequence data of the learning behavior in practice of the subsequent learning experiment combination after completing the experiment in the experiment module;
S233: defining as whether a similarity between the above two data meets a validation threshold or not, carrying on step S234 when the similarity meets the validation threshold, otherwise carrying on step S235;
S234: taking the subsequent learning experiment combination as the recommended learning combination;
S235: confirming the experiment to be a failure, carrying on the experiment and validation once again in step S213 by generating a new subsequent learning experiment combination.
In addition, whether the validation is successful or not, the invention can reenter the exploration module and the experiment module at any time to recalculate the learning mode, the presumptive learning mode and the subsequent learning experiment combination, and carry on experiment and validation.
In order to properly recommend the recommended learning combination to other users, in the present invention, after step S23, the following steps are performed:
S24: using the optimum practice module to recommend the recommended learning combination to other users who need the same or similar presumptive learning mode, so that the plural audio-visual knowledge content from multiple sources can be optimized for cohesive learning; the way of the recommended learning combination recommended by the optimum practice module including the following steps of:
S241: using the optimum practice module to convert the time-sequential learning behavior sequence data and knowledge content information of users to a series of variables;
S242: taking each user as a node to calculate clustering distances between the series of variables of each user and using this as a clustering benchmark, and classifying each user who meets thresholds of different clustering benchmarks into at least one clustering group, or the user whose average distance of one of the clustering distances is the smallest in each of the at least one clustering group being further searched as a group opinion leader;
S243: using frequent pattern mining in the optimum practice module to calculate non-clustered user who is most in line with a browsing recommended learning mode under each of the at least one knowledge cluster, and through one of recommendation, matching, competition mechanisms, or any combination of the above two or more to make the non-clustered user approach one of the at least one clustering group.
As mentioned above, the plural audio-visual knowledge content from multiple sources can produce the optimal cohesive learning mode of various learning themes without screening through a lot of manpower. The recommended learning combination can be recommended to other users with the same or similar presumptive learning mode, so that the plural audio-visual knowledge content from multiple sources can be optimized for cohesive learning. Therefore, the invention solves the problem that the traditional audio-visual knowledge content needs a lot of manpower to organize, and is more difficult to connect with the plural audio-visual knowledge content of formal and systematic education and training.
The present application is based on a provisional application Ser. No. 62/718,207 filed on Aug. 13, 2018 which is hereby incorporated by reference.
| Number | Date | Country | |
|---|---|---|---|
| 62718207 | Aug 2018 | US |
