Courseware content analysis method and device, equipment and storage medium

Abstract

The invention discloses a courseware content analysis method and device, equipment and a storage medium. The method comprises the following steps: acquiring a courseware picture; detecting each frame of courseware picture, and detecting a title and a text; inputting the detected title and text of each frame into a pre-trained character recognition model, and recognizing a title keyword set and a text keyword set of each frame of courseware picture; summarizing the title keyword set and the text keyword set of all frames of courseware pictures to obtain a courseware keyword set; obtaining a knowledge point set of a course corresponding to the courseware; obtaining a courseware integrity analysis result according to the matching-union ratio; and obtaining a courseware content analysis result according to the courseware integrity analysis result. According to the method, the teaching courseware content is analyzed and evaluated according to an objective specification standard, so that a scientific and objective evaluation conclusion is obtained.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and more specifically, to a courseware content analysis method, device, equipment and storage medium. Background Art

[0002] In the educational scenario, teaching courseware is a teaching aid that can assist teachers in teaching and explaining in class. Teaching courseware is designed to make teachers' teaching more vivid and intuitive, and help students better understand and master knowledge. At the same time, it can also be used as materials for students to review and self-study.

[0003] The analysis and evaluation of teaching courseware is also one of the most basic and core contents of education and teaching quality assurance, and is also an important part of education evaluation. At present, the current diagnosis of the teacher's courseware level depends only on people, which is highly subjective. The evaluation standards of each diagnostician are different, and it is difficult to ensure the scientificity and objectivity of the diagnosis. In addition, the current analysis and evaluation methods are aimed at the courseware outline itself, not the pictures presented by the courseware, and the evaluation results are also somewhat different from the actual feelings. Summary of the invention

[0004] In order to overcome at least one of the above-mentioned defects of the prior art, the present invention provides a courseware content analysis method, device, equipment and storage medium. The technical solution adopted by the present invention is as follows.

[0005] In a first aspect, the present invention provides a courseware content analysis method, comprising:

[0006] Obtaining a courseware picture, wherein the courseware picture is obtained by converting courseware data into image type data;

[0007] Detect each frame of the courseware image to detect the title and text;

[0008] Input the title and text detected in each frame into the pre-trained text recognition model to identify the title keyword set and text keyword set of each frame of the courseware image;

[0009] Summarize the title keyword set and the text keyword set of all the courseware images to obtain a courseware keyword set, wherein only one of each keyword is retained in the courseware keyword set without duplication;

[0010] Obtaining a set of knowledge points of the course corresponding to the courseware;

[0011] Obtaining a courseware completeness analysis result based on an intersection-and-union ratio of the courseware keyword set and the course knowledge point set;

[0012] Based on the courseware completeness analysis results, the courseware content analysis results are obtained.

[0013] In one embodiment, the process of inputting the title and text detected from each frame into a pre-trained text recognition model to identify the title keyword set and the text keyword set of each frame of the courseware image includes:

[0014] Input the title and text detected from each frame of the courseware image into the text extraction model to identify the title content and text content of each frame of the courseware image;

[0015] Use the pre-trained word segmentation model to perform text segmentation on the title content and the text content, and filter out meaningless words to obtain the word segmentation results of the title content and the text content;

[0016] Using the keyword extraction model, keywords are extracted from the word segmentation results of the title content of each frame of the courseware image and the word segmentation results of the text content;

[0017] The title keywords and text keywords extracted from each frame of the courseware are summarized to obtain the title keyword set and text keyword set of the courseware.

[0018] In one embodiment, it further includes:

[0019] The title keywords and text keywords of a single-frame courseware image are combined into an intra-frame keyword comparison group, and all intra-frame keyword comparison groups are combined into an intra-frame keyword comparison group set CompSet_A;

[0020] The comparison group set CompSet_A is input into the phrase logic relationship discrimination model to obtain the logic relationship between the title keywords and the text keywords in each frame; wherein the phrase logic relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on statistical learning or deep learning methods, and regresses the connection relationship into several pre-set logical relationship categories;

[0021] Counting the proportion R_21 of the number of frames determined to be a preset target relationship category in the entire set, wherein the preset target relationship category is one or more of the several logical relationship categories;

[0022] According to R_21, the logic analysis result within the courseware frame is obtained;

[0023] The process of obtaining the courseware content analysis result is also based on the logical analysis result within the courseware frame.

[0024] In one embodiment, the several logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others;

[0025] The preset target relationship categories include: parallel or synonymous, and subordinate.

[0026] In one embodiment, it further includes:

[0027] In the courseware picture, the keywords of the first frame to be analyzed and the keywords of the next frame of the first frame to be analyzed are taken to form an inter-frame keyword comparison group;

[0028] Input the inter-frame keyword comparison group into a phrase logical relationship discrimination model to obtain the logical relationship of the inter-frame keyword comparison group; wherein the phrase logical relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on a statistical learning or deep learning method, and regresses the connection relationship into several pre-set logical relationship categories, wherein the several logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others;

[0029] If the logical relationship between the first frame to be analyzed and the next frame of the first frame to be analyzed is subordinate or causal, then the next frame of the first frame to be analyzed is added to the content-related frame group of the first frame to be analyzed, and then the logical relationship between the keywords of the first frame to be analyzed and the keywords of the frames after the next frame of the first frame to be analyzed is continuously determined in the order of the frames until the kth frame whose logical relationship is not subordinate or causal;

[0030] Add the logical relationship between the first frame to be analyzed and the kth frame to the core content frame set CoreSet_B;

[0031] The kth frame is used as the start of the next comparison process until all courseware images are traversed to obtain a core content frame set CoreSet_B and several content-related frame groups;

[0032] According to the number of parallel categories, the number of content-related frame groups, and the proportion of content-related frames in the core content frame set CoreSet_B, the courseware structure analysis results are obtained;

[0033] The process of obtaining the courseware content analysis results is also based on the courseware structure analysis results.

[0034] In one embodiment, the process of obtaining the courseware structure analysis result according to the number of parallel categories, the number of content-related frame groups, and the proportion of content-related frames in the core content frame set CoreSet_B includes:

[0035] The calculation formula for obtaining the deviation from the optimal value is:

[0036] F_score = 1-(|x-x_best|) / x_var, where x is the actual statistical value, x_best is the best value, and x_var is the maximum deviation value;

[0037] Obtaining a first optimal value and a first maximum deviation value for the number of parallel categories, and obtaining a first deviation optimal value for the number of parallel categories based on a calculation formula for the deviation optimal value and according to the actual number of parallel categories, the first optimal value, and the first maximum deviation value;

[0038] Obtaining a second optimal value and a second maximum deviation value of the number of content-associated frame groups, and obtaining a second optimal deviation value of the content-associated frame groups based on a calculation formula for the deviation from the optimal value and according to the actual number of content-associated frame groups, the second optimal value, and the second maximum deviation value;

[0039] Obtaining a third optimal value and a second maximum deviation value of the content-associated frame quantity ratio, and obtaining a third deviation optimal value of the content-associated frame quantity ratio based on a calculation formula of the deviation optimal value and according to an actual content-associated frame quantity ratio;

[0040] The courseware structure analysis results are obtained based on the preset weight coefficients corresponding to the first, second and third deviations from the optimal value.

[0041] In one embodiment, it further includes:

[0042] Get the subject keywords of the courseware;

[0043] Input the subject keywords of the courseware and the keywords of each core content frame into the phrase logic relationship discrimination model to obtain the logical relationship between the subject keywords and the keywords of each core content frame; wherein the core content frame is the courseware picture involved in the core content frame set CoreSet_B;

[0044] The number of frames that were identified as belonging to the subordinate relationship category was counted as R_23 in the entire set;

[0045] According to the R_23, the results of the thematic coherence analysis were obtained;

[0046] The process of obtaining the courseware content analysis results is also based on the topic coherence analysis results.

[0047] In a second aspect, the present invention provides a courseware content analysis device, characterized in that it includes:

[0048] An acquisition module is used to acquire a courseware image, wherein the courseware image is obtained by converting the courseware data into image type data, and acquire a knowledge point set of the course corresponding to the courseware;

[0049] A module is selected to detect each frame of the courseware image and detect the title and text;

[0050] The recognition module is used to input the title and text detected in each frame into the pre-trained text recognition model to recognize the title keyword set and the text keyword set of each frame of the courseware image;

[0051] The first analysis module is used to obtain a courseware completeness analysis result based on the intersection and union ratio of the courseware keyword set and the course knowledge point set;

[0052] The overall analysis module is used to obtain the courseware content analysis results based on the courseware completeness analysis results.

[0053] In a third aspect, the present invention provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-mentioned methods when executing the program.

[0054] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method of any of the above-mentioned embodiments.

[0055] In the present invention, by analyzing the pictures presented in the courseware and comparing them with the keywords of the knowledge points, the completeness of the courseware content is obtained, and through the phrase logical relationship discrimination model, the logical structure within and between frames is obtained. The teaching courseware content is analyzed and evaluated with objective normative standards, thereby obtaining a scientific and objective evaluation conclusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a schematic diagram of the overall process of an implementation method of Example 1 of the present invention.

[0057] Figure 2 It is a schematic diagram of the overall process of another implementation method of the first embodiment of the present invention.

[0058] Figure 3 It is a schematic diagram of the overall process of another implementation method of the first embodiment of the present invention.

[0059] Figure 4 It is a schematic diagram of the overall process of yet another implementation method of the first embodiment of the present invention.

[0060] Figure 5 It is a schematic diagram of the overall structure of the second embodiment of the present invention. DETAILED DESCRIPTION

[0061] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0062] It should be noted that the terms "first\second\..." involved in the embodiments of the present invention are only used to distinguish similar objects, and do not represent a specific order for the objects. It is understandable that the specific order or sequence of "first\second\..." can be interchanged where permitted. It should be understood that the objects distinguished by "first\second\..." can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein.

[0063] Embodiment 1

[0064] See also Figure 1 , Figure 1 1 is a flow chart of a courseware content analysis method provided in Example 1 of the present invention, the method comprising step S110, step S120, step S130, step S140, step S150, step S160 and step S170. It should be noted that step S110, step S120, step S130, step S140, step S150, step S160 and step S170 are only reference numerals for clarifying the embodiment and the appendix. Figure 1 The corresponding relationship does not limit the order of the steps in this embodiment.

[0065] Step S110, obtaining a courseware picture, wherein the courseware picture is obtained by converting the courseware data into image type data;

[0066] Step S120, detecting each frame of the courseware image to detect the title and the text;

[0067] Step S130, input the title and text detected in each frame into a pre-trained text recognition model to identify the title keyword set KW_title and the text keyword set KW_txt of each frame of the courseware image;

[0068] Step S140, summarizing the title keyword set KW_title and the text keyword set KW_txt of all the courseware images to obtain a courseware keyword set KW_list, wherein only one of each keyword is retained in KW_list without duplication;

[0069] Step S150, obtaining a knowledge point set KW_set of the course corresponding to the courseware;

[0070] Step S160, obtaining a courseware completeness analysis result according to the intersection-and-union ratio of the KW_list and the KW_set;

[0071] Step S170, obtaining a courseware content analysis result based on the courseware completeness analysis result.

[0072] First, a courseware page data stream is input, which can be a PPT or PDF file, or a video about the courseware, and the courseware page data stream is processed to convert the courseware data into image type data to obtain courseware pictures frame by frame. In step S110, these courseware pictures frame by frame are obtained.

[0073] The input processed in this method is of image type, so it can support the analysis of courseware files of different formats. Although the PPT format files commonly used in courseware can obtain PPT content by using a specific PPT reading tool, it cannot process courseware files in PDF format or image format, is not generalizable, and has limitations on usage scenarios. In addition, during the teaching process, the courseware that students see is presented in the form of slides, and each slide can show a frame of picture. The analysis and evaluation of pictures in this method is more in line with students' actual feelings.

[0074] After obtaining the courseware image, step S120 detects each frame of the courseware image to detect the title and the text. For the specific method of detecting the title and the text in the courseware image, the courseware image can be passed into the courseware content structure detection module. The courseware content structure detection module is a module that uses image features to output the preset category and target frame position in the image. The courseware content structure detection module detects whether there is a preset category target in the input image. If the image has a target category, it outputs its position in the image and the target category. For example, the preset content category targets include: title, text.

[0075] The specific implementation methods of the courseware content structure detection module may include the following two:

[0076] Implementation method 1: Treat the preset category target detection task as a unified target detection task, use a network model trained by supervised learning method based on neural network, the network training data is training pictures with manual annotations, and the network parameter adjustment is guided by two tasks, one is the classification task of target category, and the other is the regression task of target position coordinates.

[0077] Implementation method 2: Detect each type of target separately according to its image features. Titles and text are classified as text detection. You can use the OCR detection method, and then perform empirical classification based on the size and spatial position of the text box.

[0078] After the title and text are detected, the title and text detected in each frame are input into the pre-trained text recognition model in step S130 to identify the title keyword set KW_title and the text keyword set KW_txt of each frame of the courseware image. After the text in the title and text area is identified in step S130, various keyword extraction models can be used to extract keywords. The keyword extraction model here usually uses the following technologies: 1. Statistical method: determine keywords by counting word frequency (TF-IDF); 2. Dictionary-based method: use a predefined dictionary to identify keywords, such as using a stop word list to exclude common words; 3. Machine learning-based method: train the model to identify keywords, such as using a support vector machine (SVM) or a random forest; 4. Deep learning-based method: use a neural network model, such as a convolutional neural network (CNN) or a recurrent neural network (RNN) to extract keywords.

[0079] In one embodiment, if Figure 2 As shown, the process of step S130 includes:

[0080] Step S210, inputting the title and text detected from each frame of courseware image into a text extraction model to identify the title content and text content of each frame of courseware image;

[0081] Step S220, using the pre-trained word segmentation model, performing text segmentation on the title content and the text content, and filtering out meaningless words, to obtain the word segmentation results of the title content and the text content;

[0082] Step S230, using the keyword extraction model, extracting keywords from the word segmentation results of the title content of each frame of the courseware image and the word segmentation results of the text content;

[0083] Step S240, summarizing the title keywords and text keywords extracted from each frame of the courseware image to obtain the courseware title keyword set KW_title and text keyword set KW_txt.

[0084] In step S210, the title text and the body text are extracted, such as by using OCR (Optical Character Recognition) technology to extract the title text and the body text, and identify the text content T_title in the title area and the text content T_txt in the body area.

[0085] In step S220, using a pre-trained word segmentation model, the text content T_title in the title area and the text content T_txt in the body area are segmented, and meaningless single words such as "de", "le", "zai", etc. are filtered out. The word segmentation model is a process of splitting a continuous text string into individual words or sub-words. It can use methods based on statistical learning, such as the Hidden Markov Model, or the Transformer model unsupervised trained on a large corpus.

[0086] In step S230, using a keyword extraction model, the keyword extraction is performed on the word segmentation results of the title content and the body content to obtain the most relevant vocabulary of the title text and the body text, as the knowledge point vocabulary of this frame of courseware picture. Keyword extraction is to select the most representative vocabulary from a series of vocabulary to summarize the main content of the text. It can be based on statistical methods, such as using the TF-IDF (Term Frequency-Inverse Document Frequency) method by calculating the product of the frequency of the vocabulary in the document (TF) and the inverse document frequency (IDF) in the entire document collection to measure the importance of the vocabulary; it can also use graph-based methods, such as TextRank, regarding the vocabulary as nodes in the graph, and the connections between nodes represent the co-occurrence relationship between words, and extracting keywords by calculating the PageRank value of the nodes in the graph.

[0087] After extracting the title keywords and body keywords of each frame, step S240 summarizes them to obtain the title keyword set KW_title and the body keyword set KW_txt of this courseware.

[0088] After step S130, after extracting the keywords, it enters the stage of analyzing the integrity of the courseware content. In step S140, first, the obtained keywords are integrated, and the title keyword set KW_title and the body keyword set KW_txt of all frames of courseware pictures are summarized into a preliminary knowledge point set KW_list; then, a vocabulary filter is performed on the preliminary knowledge point set KW_list to obtain a courseware keyword set KW_list of a unique term set. At this time, each keyword only retains one and there is no duplication.

[0089] In step S150, the knowledge point set KW_set of the course that the courseware should correspond to is obtained. Usually, there is a knowledge point set of the course in the course knowledge point database. The course knowledge point database is a course keyword set library that is imported in advance and covers all grades, subjects, and textbook chapters. It can be manually filled in by teachers of various disciplines, or it can be extracted using unsupervised text keywords and then corrected by teachers of various disciplines. When it is necessary to obtain the knowledge point set of the course, the courseware course noun is searched and input in the course knowledge point database to obtain the corresponding knowledge point set KW_set.

[0090] In step S160, the overlap between the courseware keyword set KW_list and the course knowledge point set KW_set is compared. Here, the jaccard similarity is used, that is, the intersection over union (IoU) of the two sets is compared. It is an indicator used to measure the similarity of the two sets and can quantify the overlap between the two sets.

[0091] J(A,B)=|A∩B| / |A∪B|;

[0092] Among them, |A∩B| represents the size of the intersection of set A and set B, and |A∪B| represents the size of the union of set A and set B.

[0093] At this time, the courseware completeness analysis result S1 is obtained based on J(A,B).

[0094] In step S170, the courseware completeness analysis result is used as one of the influencing factors to obtain the courseware content analysis result.

[0095] The above description is about taking the courseware completeness analysis result as one of the influencing factors to obtain the courseware content analysis result. In one embodiment of the present invention, the courseware intra-frame logic analysis result is also taken as one of the influencing factors to obtain the courseware content analysis result.

[0096] Specifically, Figure 3 As shown, the process of obtaining the logic analysis result within the courseware frame includes: step S310, step S320, step S330 and step S340.

[0097] Step S310, grouping the title keywords and text keywords of a single-frame courseware image into an intra-frame keyword comparison group, and combining all intra-frame keyword comparison groups into an intra-frame keyword comparison group set CompSet_A;

[0098] Step S320, inputting the comparison group set CompSet_A into a phrase logic relationship discrimination model to obtain the logic relationship between the title keywords and the text keywords in each frame; wherein the phrase logic relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on statistical learning or deep learning methods, and regresses the connection relationship into several pre-set logical relationship categories;

[0099] Step S330, counting the proportion R_21 of the number of frames determined to be of a preset target relationship category in the entire set, wherein the preset target relationship category is one or more of the several logical relationship categories;

[0100] Step S340, obtaining the logic analysis result within the courseware frame according to R_21.

[0101] In step S310, the title keywords and text keywords of a single courseware frame are grouped into an intra-frame keyword comparison group, and all intra-frame keyword comparison groups are grouped into an intra-frame keyword comparison group set CompSet_A. The intra-courseware frame logic analysis determines whether the internal logic connection of the frame content is close by determining the correlation between the title and text keywords.

[0102] In step S320, the intra-frame keyword comparison group set CompSet_A is input into the phrase logical relationship discrimination model to obtain the logical relationship between the title and the text keywords in each frame.

[0103] The phrase logical relationship discrimination model is a method for judging the logical relationship between two words. It mines the connection between nouns in large-scale natural language data based on statistical learning or deep learning methods, and regresses the connection to the six set logical relationship categories. Preferably, the six logical relationships include: parallel or synonymous, contrast or antonym, subordination, causality, progression, and others. Among them, the input form of keywords should be a feature representation through text conversion, such as word vector (word2vec) or word embedding (embedding); the learning model can adopt a model based on statistical learning, such as support vector machine (SVM), random forest, etc., or a deep learning model, such as a network model based on Transformer structure, etc., to extract the relational features of the input vector for modeling; the relational features of the input vector can be the high-dimensional space distance of the word vector, the collinear word frequency of the phrase, the contextual semantic embedding, etc.

[0104] In step S330, the proportion of the number of frames determined as the preset target relationship category in the entire set is counted R_21, and the preset target relationship category is one or more of the previous logical relationship categories. Preferably, in the case where the six logical relationships are determined in the previous, the preset target relationship categories include: parallel or synonymous, and subordinate. Preferably, the position of the courseware frame whose logical relationship in the frame is determined to be a non-subordinate relationship is recorded to facilitate later modification.

[0105] Finally, step S340 obtains the result of the logic analysis within the courseware frame based on the proportion R_21.

[0106] In addition, in one embodiment of the present invention, the courseware structure analysis result regarding the logic analysis between courseware frames is used as one of the influencing factors to obtain the courseware content analysis result.

[0107] Specifically, Figure 4 As shown, the process of obtaining the courseware structure analysis result includes: step S410, step S420, step S430, step S440, step S450 and step S460.

[0108] Step S410, in the courseware picture, taking the keywords of the first frame to be analyzed and the keywords of the next frame of the first frame to be analyzed to form an inter-frame keyword comparison group;

[0109] Step S420, inputting the inter-frame keyword comparison group into a phrase logical relationship discrimination model to obtain the logical relationship of the inter-frame keyword comparison group; wherein the phrase logical relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on a statistical learning or deep learning method, and regresses the connection relationship into a plurality of pre-set logical relationship categories, wherein the plurality of logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others;

[0110] Step S430, if the logical relationship between the first frame to be analyzed and the next frame of the first frame to be analyzed is subordinate or causal, then the next frame of the first frame to be analyzed is added to the content-related frame group of the first frame to be analyzed, and then the logical relationship between the keyword of the first frame to be analyzed and the keyword of the frame after the next frame of the first frame to be analyzed is continuously determined in the order of the frames until the kth frame whose logical relationship is not subordinate or causal;

[0111] Step S440, adding the logical relationship between the first frame to be analyzed and the kth frame to the core content frame set CoreSet_B;

[0112] Step S450, starting the next comparison process from the kth frame until all courseware images are traversed to obtain a core content frame set CoreSet_B and a number of content-related frame groups;

[0113] Step S460, obtaining the courseware structure analysis result according to the number of parallel categories, the number of content-related frame groups, and the proportion of content-related frames in the core content frame set CoreSet_B.

[0114] Obtaining the logical relationship between the entire courseware frames and identifying the relationship between keywords between frames helps to understand the logical flow of the courseware content. First, identify the core content pages and secondary content pages in the courseware. The core content is usually the theme or main concept of the courseware, while the secondary content page may be the details that support, explain or expand the core content. For example, the main theme of the courseware contains several first-level branches. The homepage slide that introduces each first-level branch is the core content page, and the following slides that explain this first-level branch are secondary content pages.

[0115] The first frame of the courseware image to be analyzed can be manually specified, usually the first frame with actual content except for the cover and directory. In addition, in the scenario of automatic acquisition and analysis, because the first frame is generally assumed to be the courseware cover frame, statistics can be started from the second frame content. The keywords of the second frame F_1 and the keywords of the next frame (i.e., the third frame) F_2 are obtained to form an inter-frame keyword comparison group, and the phrase logical relationship discrimination model is input to obtain the logical relationship between the two frames, wherein the phrase logical relationship discrimination model can adopt the phrase logical relationship discrimination model of the previous implementation method. If the third frame is in a subordinate relationship or a causal relationship with the second frame, F_2 is added to the content-related frame group SubSet_1 of F_1, and then the logical relationship between the keywords of the first frame F_1 and the keywords of the next frame (i.e., the fourth frame) F_3 is continued to be judged until the keywords of F_1 and the keywords of the next frame F_k are not in a subordinate relationship or a causal relationship. The reason for doing this is as follows: As mentioned above, the courseware contains core content pages and secondary content pages. The secondary content page may be the details that support, explain or expand the core content. Therefore, when there is a subordinate relationship or causal relationship, it can be judged that the page is a secondary content page about the previous page. When it is judged that the keywords of F_1 and the keywords of the next frame F_k are not in a subordinate relationship or causal relationship, F_k is used as the beginning of the next comparison process until the entire courseware frame is traversed. When the keywords of F_1 and the keywords of the next frame F_k are not in a subordinate relationship or causal relationship, it proves that the next first-level branch has been entered, and this frame F_k is the core content page of the next first-level branch. At the same time, the logical relationship between F_1 and F_k is added to the core content frame set CoreSet_B. This core content frame set CoreSet_B is the set of logical relationships between the homepage of this first-level branch and the homepage of the next-level branch.

[0116] After the above frame comparison process, a core content frame set CoreSet_B and several content-related frame groups SubSet_i will be obtained.

[0117] According to the number of parallel categories, the number of content-related frame groups, and the proportion of content-related frames (calculated by the number of each content-related frame group and the total number of frames of the courseware pictures) in the core content frame set CoreSet_B, the type of the courseware logical structure is determined, and the courseware structure analysis results are obtained, such as linear structure, hierarchical structure and other structures. Among them, the linear structure is that the explanation between each first-level branch has a focus, some are explained more, and some are explained less. The hierarchical structure is that the amount of explanation between each first-level branch is similar.

[0118] Specifically, the score of the courseware logical structure is calculated based on the number N1 of parallel categories in the core content frame set CoreSet_B, the number N2 of content-related frame groups, and the proportion R_NC of the number of content-related frames in the courseware frame set.

[0119] These three statistical indicators do not follow the principle that the larger the better. Here, we will use the deviation from the optimal value to evaluate. The score is calculated based on the degree of deviation between the actual statistical value and the ideal middle value.

[0120] Number of parallel categories: Set the optimal value to 6 and the maximum deviation value to 3, which is used to count the number of knowledge points in the courseware. Too many or too few will affect the logical structure.

[0121] Number of content-associated frame groups: Set the optimal value to 5 groups, and the maximum deviation value to 2, which is used to count the number of knowledge points detailed in the courseware and reflect the hierarchical distribution of the content.

[0122] The proportion of content-related frames: The optimal value is set to 60%-80%, and the maximum deviation value is 20%, indicating that the proportion of knowledge points explained in detail should not be overly concentrated or dispersed.

[0123] The calculation formula for the deviation from the optimal value is:

[0124] F_score=1-(|x-x_best|) / x_var

[0125] Among them, x is the actual statistical value, x_best is the ideal optimal value, and x_var is the maximum deviation value;

[0126] Comprehensive evaluation score

[0127] C_22=w_N1*F_score(N1)+w_N2*F_score(N2)+w_R*F_score(R_NC)Wherein, w_N1, w_N2, w_R are factor weight coefficients, and the experimental recommended values ​​are: 0.3, 0.3, 0.4;

[0128] This method is a structural evaluation that neither pursues extreme values ​​nor is too conservative. The closer the score is to 100, the closer the logical structure of the courseware is to the ideal state, and the more conducive it is for students to understand and learn.

[0129] In addition, in one embodiment of the present invention, the result of the subject coherence analysis is used as one of the influencing factors to obtain the result of the courseware content analysis.

[0130] Specifically, Figure 5 As shown, the process of obtaining the topic coherence analysis result includes: step S510, step S520, step S530 and step S540.

[0131] Step S510, obtaining the subject keywords of the courseware;

[0132] Step S520, inputting the subject keywords of the courseware and the keywords of each core content frame into the phrase logical relationship discrimination model to obtain the logical relationship between the subject keywords and the keywords of each core content frame; wherein the core content frame is the courseware picture involved in the core content frame set CoreSet_B;

[0133] Step S530, counting the proportion of the number of frames that are determined to be in the subordinate relationship category in the entire set, R_23;

[0134] Step S540, obtaining the result of the topic coherence analysis according to R_23.

[0135] Step S510 obtains the subject keywords of the courseware. The subject keywords of the courseware can be manually input, or, in the scenario of automatic acquisition and analysis, the first frame is defaulted to be the courseware cover frame, and its keywords are used as the subject keywords of the courseware.

[0136] In this embodiment, the core content page in the previous embodiment is used. As mentioned above, the core content frame set CoreSet_B is the set of logical relationships between the first-level branch homepage and the next-level branch homepage. Therefore, the courseware pictures involved in the core content frame set CoreSet_B are all core content pages, so that the core content frame keyword set can be obtained. Through the phrase logical relationship discrimination model, the relationship between each core content frame keyword and the courseware theme keyword is discriminated, and the proportion of the core content frame and the courseware theme in a subordinate relationship R_23 is counted to determine whether the content of this courseware is presented around this theme keyword.

[0137] In this method, various factors that affect the evaluation of courseware content analysis results include: courseware completeness analysis results, courseware intra-frame logic analysis results, courseware structure analysis results, and theme coherence analysis results. These factors can be multiplied by different weight coefficients to ultimately obtain courseware content analysis results.

[0138] In this method, the images presented in the courseware are analyzed and compared with the keywords of the knowledge points to obtain the completeness of the courseware content. The logical structure within and between frames is obtained through the phrase logical relationship discrimination model. The teaching courseware content is analyzed and evaluated with objective normative standards, thus obtaining a scientific and objective evaluation conclusion.

[0139] Embodiment 2

[0140] Corresponding to the method of embodiment 1, as Figure 5As shown, the present invention also provides a courseware content analysis device, comprising: an acquisition module 510, a selection module 520, an identification module 530, a first analysis module 540 and a total analysis module 550;

[0141] The acquisition module 510 is used to acquire a courseware image, wherein the courseware image is obtained by converting the courseware data into image type data, and acquire a knowledge point set KW_set of the course corresponding to the courseware;

[0142] A selection module 520 is used to detect each frame of the courseware image and detect the title and the text;

[0143] The recognition module 530 is used to input the title and text detected in each frame into a pre-trained text recognition model to recognize the title keyword set KW_title and the text keyword set KW_txt of each frame of the courseware image;

[0144] The first analysis module 540 is used to obtain a courseware completeness analysis result according to the intersection-and-union ratio of the KW_list and the KW_set;

[0145] The overall analysis module 550 is used to obtain the courseware content analysis result based on the courseware completeness analysis result.

[0146] In one embodiment, the recognition module performs the process of inputting the title and text detected from each frame into a pre-trained text recognition model to recognize the title keyword set KW_title and the text keyword set KW_txt of each frame of the courseware image, including:

[0147] Input the title and text detected from each frame of the courseware image into the text extraction model to identify the title content and text content of each frame of the courseware image;

[0148] Use the pre-trained word segmentation model to perform text segmentation on the title content and the text content, and filter out meaningless words to obtain the word segmentation results of the title content and the text content;

[0149] Using the keyword extraction model, keywords are extracted from the word segmentation results of the title content of each frame of the courseware image and the word segmentation results of the text content;

[0150] The title keywords and text keywords extracted from each frame of the courseware images are summarized to obtain the courseware title keyword set KW_title and text keyword set KW_txt.

[0151] In one embodiment, it further includes: a second analysis module;

[0152] The second analysis module is used to group the title keywords and text keywords of a single-frame courseware image into an intra-frame keyword comparison group, and combine all the intra-frame keyword comparison groups into an intra-frame keyword comparison group set CompSet_A;

[0153] The comparison group set CompSet_A is input into the phrase logic relationship discrimination model to obtain the logic relationship between the title keywords and the text keywords in each frame; wherein the phrase logic relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on statistical learning or deep learning methods, and regresses the connection relationship into several pre-set logical relationship categories;

[0154] Counting the proportion R_21 of the number of frames determined to be of a preset target relationship category in the entire set, wherein the preset target relationship category is one or more of the several logical relationship categories;

[0155] According to R_21, the logic analysis result within the courseware frame is obtained;

[0156] The overall analysis module is also used to obtain courseware content analysis results based on the logic analysis results within the courseware frame.

[0157] In one embodiment, the several logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others;

[0158] The preset target relationship categories include: parallel or synonymous, and subordinate.

[0159] In one embodiment, it further includes: a third analysis module;

[0160] The third analysis module is used to take the keywords of the first frame to be analyzed and the keywords of the next frame of the first frame to be analyzed in the courseware picture to form an inter-frame keyword comparison group;

[0161] Input the inter-frame keyword comparison group into a phrase logical relationship discrimination model to obtain the logical relationship of the inter-frame keyword comparison group; wherein the phrase logical relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on a statistical learning or deep learning method, and regresses the connection relationship into several pre-set logical relationship categories, wherein the several logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others;

[0162] If the logical relationship between the first frame to be analyzed and the next frame of the first frame to be analyzed is subordinate or causal, then the next frame of the first frame to be analyzed is added to the content-related frame group of the first frame to be analyzed, and then the logical relationship between the keywords of the first frame to be analyzed and the keywords of the frames after the next frame of the first frame to be analyzed is continuously determined in the order of the frames until the kth frame whose logical relationship is not subordinate or causal;

[0163] Add the logical relationship between the first frame to be analyzed and the kth frame to the core content frame set CoreSet_B;

[0164] The kth frame is used as the start of the next comparison process until all courseware images are traversed to obtain a core content frame set CoreSet_B and several content-related frame groups;

[0165] According to the number of parallel categories in the core content frame set CoreSet_B, the number of content-related frame groups, the number of each content-related frame group and the total number of courseware pictures, the courseware structure analysis results are obtained;

[0166] The overall analysis module is also used to obtain courseware content analysis results based on the courseware structure analysis results.

[0167] In one embodiment, it further includes: a fourth analysis module;

[0168] The fourth analysis module is used to obtain the subject keywords of the courseware;

[0169] Input the subject keywords of the courseware and the keywords of each core content frame into the phrase logic relationship discrimination model to obtain the logical relationship between the subject keywords and the keywords of each core content frame; wherein the core content frame is the courseware picture involved in the core content frame set CoreSet_B;

[0170] The number of frames that were identified as belonging to the subordinate relationship category was counted as R_23 in the entire set;

[0171] According to the R_23, the results of the thematic coherence analysis were obtained;

[0172] The overall analysis module is also used to obtain courseware content analysis results based on the topic coherence analysis results.

[0173] In this device, the images presented in the courseware are analyzed and compared with the keywords of the knowledge points to obtain the completeness of the courseware content. The logical structure within and between frames is obtained through the phrase logical relationship discrimination model. The teaching courseware content is analyzed and evaluated with objective normative standards, thereby obtaining a scientific and objective evaluation conclusion.

[0174] Embodiment 3

[0175] An embodiment of the present invention further provides a storage medium on which computer instructions are stored. When the instructions are executed by a processor, the courseware content analysis method of any of the above embodiments is implemented.

[0176] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, random access memories (RAM), read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0177] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the relevant technology. The computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the methods of each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, RAM, ROM, magnetic disks or optical disks.

[0178] Corresponding to the above-mentioned computer storage medium, in one embodiment, a computer device is also provided, which includes a memory, an encoder, and a computer program stored in the memory and executable on the encoder, wherein when the encoder executes the program, it implements any one of the courseware content analysis methods in the above-mentioned embodiments.

[0179] The above-mentioned computer equipment analyzes the images presented in the courseware, compares them with the keywords of the knowledge points, and obtains the completeness of the courseware content. It uses the phrase logical relationship discrimination model to obtain the logical structure within and between frames, and analyzes and evaluates the teaching courseware content with objective and standardized standards, thereby obtaining a scientific and objective evaluation conclusion.

[0180] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0181] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A courseware content analysis method, characterized in that: include: Obtaining a courseware picture, wherein the courseware picture is obtained by converting courseware data into image type data; Detect each frame of the courseware image to detect the title and text; Input the title and text detected in each frame into the pre-trained text recognition model to identify the title keyword set and text keyword set of each frame of the courseware image; Summarize the title keyword set and the text keyword set of all the courseware images to obtain a courseware keyword set, wherein only one of each keyword is retained in the courseware keyword set without duplication; Obtaining a set of knowledge points of the course corresponding to the courseware; Obtaining a courseware completeness analysis result based on an intersection-and-union ratio of the courseware keyword set and the course knowledge point set; Based on the courseware completeness analysis results, the courseware content analysis results are obtained.

2. The courseware content analysis method according to claim 1, characterized in that: The process of inputting the title and text detected from each frame into a pre-trained text recognition model to recognize the title keyword set and the text keyword set of each frame of the courseware image includes: Input the title and text detected from each frame of the courseware image into the text extraction model to identify the title content and text content of each frame of the courseware image; Use the pre-trained word segmentation model to perform text segmentation on the title content and the text content, and filter out meaningless words to obtain the word segmentation results of the title content and the text content; Using the keyword extraction model, keywords are extracted from the word segmentation results of the title content of each frame of the courseware image and the word segmentation results of the text content; The title keywords and text keywords extracted from each frame of the courseware are summarized to obtain the title keyword set and text keyword set of the courseware.

3. The courseware content analysis method according to claim 1 or 2, characterized in that: Also includes: The title keywords and text keywords of a single-frame courseware image are combined into an intra-frame keyword comparison group, and all intra-frame keyword comparison groups are combined into an intra-frame keyword comparison group set CompSet_A; The comparison group set CompSet_A is input into the phrase logic relationship discrimination model to obtain the logic relationship between the title keywords and the text keywords in each frame; wherein the phrase logic relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on statistical learning or deep learning methods, and regresses the connection relationship into several pre-set logical relationship categories; Counting the proportion R_21 of the number of frames determined to be of a preset target relationship category in the entire set, wherein the preset target relationship category is one or more of the several logical relationship categories; According to R_21, the logic analysis result within the courseware frame is obtained; The process of obtaining the courseware content analysis result is also based on the logical analysis result within the courseware frame.

4. The courseware content analysis method according to claim 3, characterized in that: The several logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others; The preset target relationship categories include: parallel or synonymous, and subordinate.

5. The courseware content analysis method according to claim 1 or 2, characterized in that: Also includes: In the courseware picture, the keywords of the first frame to be analyzed and the keywords of the next frame of the first frame to be analyzed are taken to form an inter-frame keyword comparison group; Input the inter-frame keyword comparison group into a phrase logical relationship discrimination model to obtain the logical relationship of the inter-frame keyword comparison group; wherein the phrase logical relationship discrimination model is a model that mines the connection relationship between nouns in large-scale natural language data based on a statistical learning or deep learning method, and regresses the connection relationship into several pre-set logical relationship categories, wherein the several logical relationship categories include: parallel or synonymous, contrast or antonym, subordination, cause and effect, progression, and others; If the logical relationship between the first frame to be analyzed and the next frame of the first frame to be analyzed is subordinate or causal, then the next frame of the first frame to be analyzed is added to the content-related frame group of the first frame to be analyzed, and then the logical relationship between the keywords of the first frame to be analyzed and the keywords of the frames after the next frame of the first frame to be analyzed is continuously determined in the order of the frames until the kth frame whose logical relationship is not subordinate or causal; Add the logical relationship between the first frame to be analyzed and the kth frame to the core content frame set CoreSet_B; The kth frame is used as the start of the next comparison process until all courseware images are traversed to obtain a core content frame set CoreSet_B and several content-related frame groups; According to the number of parallel categories, the number of content-related frame groups, and the proportion of content-related frames in the core content frame set CoreSet_B, the courseware structure analysis results are obtained; The process of obtaining the courseware content analysis results is also based on the courseware structure analysis results.

6. The courseware content analysis method according to claim 5, characterized in that: The process of obtaining the courseware structure analysis result according to the number of parallel categories, the number of content-related frame groups, and the proportion of content-related frames in the core content frame set CoreSet_B includes: The calculation formula for obtaining the deviation from the optimal value is: F_score = 1-(|x-x_best|) / x_var, where x is the actual statistical value, x_best is the best value, and x_var is the maximum deviation value; Obtaining a first optimal value and a first maximum deviation value for the number of parallel categories, and obtaining a first deviation optimal value for the number of parallel categories based on a calculation formula for the deviation optimal value and according to the actual number of parallel categories, the first optimal value, and the first maximum deviation value; Obtaining a second optimal value and a second maximum deviation value of the number of content-associated frame groups, and obtaining a second optimal deviation value of the content-associated frame groups based on a calculation formula for the deviation from the optimal value and according to the actual number of content-associated frame groups, the second optimal value, and the second maximum deviation value; Obtaining a third optimal value and a second maximum deviation value of the content-associated frame quantity ratio, and obtaining a third deviation optimal value of the content-associated frame quantity ratio based on a calculation formula of the deviation optimal value and according to an actual content-associated frame quantity ratio; The courseware structure analysis results are obtained based on the preset weight coefficients corresponding to the first, second and third deviations from the optimal value.

7. The courseware content analysis method according to claim 5, characterized in that: Also includes: Get the subject keywords of the courseware; Input the subject keywords of the courseware and the keywords of each core content frame into the phrase logic relationship discrimination model to obtain the logical relationship between the subject keywords and the keywords of each core content frame; wherein the core content frame is the courseware picture involved in the core content frame set CoreSet_B; The number of frames that were identified as belonging to the subordinate relationship category was counted as R_23 in the entire set; According to the R_23, the results of the thematic coherence analysis were obtained; The process of obtaining the courseware content analysis results is also based on the topic coherence analysis results.

8. A courseware content analysis device, characterized in that: include: An acquisition module is used to acquire a courseware image, wherein the courseware image is obtained by converting the courseware data into image type data, and acquire a knowledge point set of the course corresponding to the courseware; A module is selected to detect each frame of the courseware image and detect the title and text; The recognition module is used to input the title and text detected in each frame into the pre-trained text recognition model to recognize the title keyword set and text keyword set of each frame of the courseware image; The first analysis module is used to obtain a courseware completeness analysis result based on the intersection and union ratio of the courseware keyword set and the course knowledge point set; The overall analysis module is used to obtain the courseware content analysis results based on the courseware completeness analysis results.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images