Relevance detection device and relevance detection method

Abstract

A relevance detection device according to the present invention comprises: a topic extraction unit that extracts topics handled by content from a document assigned to the content; a topic set expansion unit that adds topics similar to the extracted topics to the extracted topics to obtain an expanded topic set; and a topic range comparison unit that compares topic sets respectively handled by a plurality of pieces of content with each other to detect a topic range specifying an overlap between the plurality of topic sets.

Description

Relevance Detection Device and Relevance Detection Method

[0001] The present invention relates to a relevance detection device and a relevance detection method for detecting the relevance between educational contents provided by, for example, e-learning or the like.

[0002] In recent years, a large number of learning management systems that provide a plurality of educational contents on the same platform, such as e-learning courses, have been provided and are used in corporate training and the like.

[0003] In this type of learning management system, an overview of each educational content is presented. As a result, on the learning management system, evaluations, levels, etc. of each course are presented individually.

[0004] Nils Reimers and Iryna Gurevych: Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks, https: / / arxiv.org / abs / 1908.10084 (searched on September 13, 2024)

[0005] However, since these learning management systems do not have a method for detecting the relevance between the contents handled by the educational contents, the learners do not know the relevance between the contents.

[0006] As an existing technique, there is a method of comparing documents included in contents, for example, as in Non-Patent Document 1. This method calculates the vectors of the entire documents and calculates the similarity between the documents by comparing the vectors.

[0007] However, when this method is applied, the similarity in parts not related to education is calculated. Since the documents attached to the educational contents are generally short, it is likely to be inappropriate for calculating the similarity.

[0008] For example, when comparing a document titled "This is an introductory course to learn how to use Excel through exercises" with a document titled "This is an introductory course to learn how to use AWS through exercises," the similarity of the documents will be high even though the content is completely different from an educational standpoint (Example 1). Similarly, when comparing a document titled "Let's learn AWS from scratch!" with a document titled "This is an introductory course to learn how to use AWS through exercises," the similarity of the documents will be lower than in Example 1, even though the content is similar from an educational standpoint (Example 2).

[0009] Furthermore, existing technologies cannot demonstrate relevance to content that is not explicitly stated within the document.

[0010] Therefore, learners cannot determine which educational content to combine and how to combine them to be most effective.

[0011] This invention has been made in view of these circumstances, and aims to provide a relationship detection device and relationship detection method that can detect relationships between educational content provided, for example, through e-learning.

[0012] To achieve the above objective, a first aspect of the present invention is a relationship detection device comprising: a topic extraction unit that extracts topics dealt with by the content from a document attached to the content; a topic set expansion unit that adds similar topics to the extracted topics to obtain an expanded topic set; and a topic range comparison unit that compares the topic sets dealt with by multiple contents and detects topic ranges that clearly indicate the overlap of multiple topic sets.

[0013] A second aspect of the present invention is the relationship detection device of the first aspect, wherein the topic extraction unit extracts characteristic words from a document, compares the extracted words with educational documents and other general documents, selects words that are more frequent in educational documents from among the extracted words, creates a word vector based on the selected words, clusters similar words using the word vector, and extracts the word with the highest frequency of occurrence among the clustered words as a topic.

[0014] A third aspect of the present invention is the relationship detection device of the first aspect, wherein the topic set expansion unit vectorizes the topics of the content, compares the vectorized topics with word vectors in a pre-prepared topic dictionary, selects topics from the vectorized topics in order of similarity based on the comparison result, and adds the selected topics to the topics extracted by the topic extraction unit to obtain an expanded topic set.

[0015] A fourth aspect of the present invention is a relevance detection method implemented by a relevance detection device, wherein the processor of the relevance detection device extracts topics dealt with by the content from documents assigned to the content, adds similar topics to the extracted topics to obtain an expanded topic set, compares the multiple topic sets dealt with by multiple contents, and detects topic ranges that clearly indicate the overlap of the multiple topic sets.

[0016] According to the relationship detection device and relationship detection method of the present invention, relationships between content in a learning management system can be detected by comparing content using only a specific topic, such as education. Furthermore, by expanding the topics of the content using data on a specific topic, such as education, it becomes possible to detect relationships that are not superficially apparent.

[0017] Figure 1 is a functional block diagram showing an example configuration of a relevance detection device to which the relevance detection method according to an embodiment of the present invention is applied. Figure 2 is a conceptual diagram illustrating an example of topic extraction processing. Figure 3 is a conceptual diagram illustrating an example of topic set expansion processing. Figure 4 is a conceptual diagram illustrating an example of topic range comparison processing. Figure 5 is a simplified diagram showing the hardware configuration of a computer. Figure 6 is a flowchart illustrating an example of operation of a relevance detection device to which the relevance detection method according to an embodiment of the present invention is applied. Figure 7 is a flowchart illustrating the flow of topic extraction processing by the topic extraction unit. Figure 8 is a schematic diagram illustrating an example of topic extraction processing by the topic extraction unit. Figure 9 is a flowchart illustrating the flow of topic set expansion processing by the topic set expansion unit. Figure 10 is a flowchart illustrating the flow of topic range comparison processing by the topic range comparison unit. Figure 11 is a Venn diagram showing an example of selecting items with exact word matches as common topics. Figure 12 is an example of a Venn diagram used for determining common topics.

[0018] Embodiments of the present invention will be described below with reference to the drawings. The drawings are schematic or conceptual, and the proportions of each part and the proportions between parts are not necessarily the same as those of reality. Furthermore, even when representing the same part, the dimensions and proportions may be represented differently in the drawings. In this specification and in each drawing, elements similar to those described in previously shown drawings are denoted by the same reference numerals, and detailed explanations and redundant explanations are omitted as appropriate.

[0019] Figure 1 is a functional block diagram showing an example of the configuration of a relationship detection device to which the relationship detection method according to an embodiment of the present invention is applied.

[0020] In other words, the relationship detection device 10 according to the embodiment of the present invention is a device that, by comparing content in a learning management system using only specific content such as educational content, presents relationships such as overlaps and differences in content within that specific content.

[0021] The following explanation will use education as an example of specific content, but the relevance detection device 10 according to the present invention is not limited to content related to education.

[0022] If a particular piece of content is educational, the relevance detection device 10 according to the present invention extracts educational topics from documents such as summaries attached to the educational content, adds similar educational topics to the extracted topics to obtain an expanded topic set, and compares the topic sets handled by each of the multiple pieces of content to detect topic ranges that clearly indicate the overlap of these topic sets, thereby detecting the relevance between the multiple pieces of content.

[0023] To achieve this, the relevance detection device 10 includes a topic extraction unit 20, a topic set expansion unit 30, and a topic range comparison unit 40.

[0024] The topic extraction unit 20 performs topic extraction processing.

[0025] Figure 2 is a conceptual diagram illustrating an example of topic extraction processing.

[0026] The topic extraction process, as illustrated in Figure 2, is the process of extracting topics (units, specific technical terms, etc.) 104 that the content deals with from the content summary text 100 and table of contents, etc. 102 on the learning management system.

[0027] The topic set expansion unit 30 executes topic set expansion processing.

[0028] Figure 3 is a conceptual diagram illustrating an example of topic set expansion processing.

[0029] The topic set expansion process, as illustrated in Figure 3, expands the topic 104 of the content by adding similar topics 108 to it, thereby obtaining an expanded topic set 110 related to this content.

[0030] The topic range comparison unit 40 performs topic range comparison processing.

[0031] Figure 4 is a conceptual diagram illustrating an example of topic range comparison processing.

[0032] The topic range comparison process, as illustrated in Figure 4, compares extended topic sets (112(#A), 112(#B)) handled by multiple contents (for example, the contents of Course A and the contents of Course B), and detects the relationship between multiple contents by detecting topic ranges 114 that explicitly show the overlap of the multiple topic sets as a result of the comparison. This also makes it possible to identify common topics and topics specific to each content.

[0033] In comparing topic sets, using an extended topic set makes it possible to detect relationships that could not be detected using only the documents originally attached to the content.

[0034] Such a correlation detection device 10 can be implemented using a computer such as a PC.

[0035] Figure 5 is a simplified diagram showing the hardware configuration of a computer.

[0036] Computer 200 receives some kind of input from the outside, processes it, and outputs the result to the outside. Input device 220 handles the input, and output device 240 handles the output. CPU (Central Processing Unit) 210 controls the entire flow of data and processing, and performs calculations and other processing. In Figure 5, solid arrows represent the flow of data, and dashed arrows represent the flow of control.

[0037] For the computer 200 to perform processing, it needs a program that describes the processing procedure. The program contains a series of instructions and the data used by those instructions, all of which are temporarily stored in memory 230. The CPU 210 retrieves the instructions from memory 230 one by one, interprets them, and operates according to those instructions.

[0038] Therefore, the relevance detection device 10 is realized by the operation of the CPU 210 according to the program stored in the memory 230. Although not shown in FIG. 5, the computer 200 can incorporate a storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive).

[0039] Although the relevance detection device 10 is realized by a computer such as a PC as described above, it should be noted that in FIG. 1, the general configuration requirements of the computer are omitted, the configuration is simplified, and only the minimum necessary configuration for the invention is described.

[0040] Next, the operation of the relevance detection device to which the relevance detection method according to the embodiment of the present invention configured as described above is applied will be described.

[0041] FIG. 6 is a flowchart showing an operation example of the relevance detection device to which the relevance detection method according to the embodiment of the present invention is applied.

[0042] As illustrated in FIG. 6, in the relevance detection device 10, topic extraction processing is performed by the topic extraction unit 20 (S1), topic set expansion processing is performed by the topic set expansion unit 30 (S2), and topic extraction comparison processing is performed by the topic range comparison unit 40 (S4).

[0043] FIG. 7 is a flowchart showing the flow of topic extraction processing by the topic extraction unit.

[0044] FIG. 8 is a schematic diagram for explaining an example of topic extraction processing by the topic extraction unit.

[0045] That is, in the topic extraction processing (S1), the topic extraction unit 20 reads the content summary sentence, the content document such as the explanatory text 100 and the table of contents 102, etc. from the document on the platform (S11), extracts topic candidates related to the education of the content (S12), and selects the topic 104 (S13).

[0046] A first example of the topic candidate extraction processing (S12) performed by the topic extraction unit 20 will be described.

[0047] In the first example, when the list of topics is known, the topic extraction unit 20 can, for example, perform pattern matching with the words in the list and extract topic candidates based on the results of the pattern matching. Further, the topic extraction unit 20 can also calculate the similarity between the vectors of the words in the list and the vectors of the words in the document, and extract the words with high similarity as topic candidates.

[0048] Also, in the first example, when the list of topics is unknown, the topic extraction unit 20 can extract important words obtained using techniques such as TF-IDF as topic candidates. Note that TF-IDF is a known method for detecting characteristic words that appear frequently in a document but not generally frequently.

[0049] Next, a second example of the topic candidate extraction process (S12) performed by the topic extraction unit 20 will be described.

[0050] In the second example, the topic extraction unit 20 can extract education-related topics using an existing dictionary, thesaurus, etc. The topic extraction unit 20 can also create a dictionary and extract only education-related topics, for example, as exemplified below.

[0051] When creating a dictionary to extract education-related topics, the topic extraction unit 20, for example, collects documents on the platform and extracts characteristic words therefrom. Next, it compares with other general documents and education-related documents, and selects the words that are less in the former documents and more in the latter documents among the extracted words. Thereafter, word vectors are created and clustered according to the distance. Note that the word vectors may be created in the same manner as the method described in the first example, or unique vectors may be created, such as creating them from the collected documents on the platform.

[0052] Next, the details of the topic selection process (S13) performed by the topic extraction unit 20 will be described.

[0053] The topic extraction unit 20 clusters similar words using word vectors from the topic candidates. In this case, existing vectors may be used to create the word vectors, or existing models and technologies such as word2vec or BERT may be used. Alternatively, they may be created independently using document sets within the platform or other education-related document sets.

[0054] The topic extraction unit 20 then selects the word with the highest frequency of occurrence among the clustered words as the topic.

[0055] In this way, the topic extraction unit 20 performs the topic extraction process (S1).

[0056] Next, we will explain in detail the topic set expansion process (S2) performed by the topic set expansion unit 30.

[0057] Figure 9 is a flowchart showing the flow of the topic set expansion process performed by the topic set expansion unit 30.

[0058] In the topic set expansion process (S2), the topic set expansion unit 30 vectorizes the topics of each content, compares them with the word vectors in a pre-prepared topic dictionary, and based on the comparison results, selects a number of topics (where a is an arbitrary natural number) from the vectorized topics in descending order of cosine similarity, and adds the selected topics to the original topics extracted by the topic extraction unit 20. This results in an expanded topic set.

[0059] The topic dictionary can be a dictionary used in the topic extraction process (S1) to extract education-related topics. This ensures that the selected topics are education-related topics similar to each content item.

[0060] Comparisons can also be made for content with few topics or content with topics that appear infrequently.

[0061] Specifically, as shown in the flowchart in Figure 9, the topic set extension unit 30 reads the topic set GT (number of elements: N) of the content to be extended (S21), reads the education-related topic dictionary TD to be used for extension (S22), sets the count i to 0 (zero) (S23), and starts topic extension (S24). Then, from the education-related topic dictionary TD, it reads the i-th topic T of the content to be extended. ｉ Select a number of topics (TD_0 to TD_a) with a high similarity to the given topic (S25), set the count j to 0 (zero) (S26), and begin adding topics (S27). If topic TD_i does not exist in the topic set GT (S28: Yes), add topic TD_i to the topic set GT (S29). If j < a (S30: Yes), set the counter j to j + 1 (S31), and end the topic addition process (S32).

[0062] On the other hand, if topic TD_i exists in step S28 (S28: No), step S29 is skipped and the process proceeds to step S30. If j < a (S30: Yes), the counter j = j + 1 (S31), and the topic addition is completed (S32). After step S32, and in step 30, if j < a is not true (S30: No), and i < N (S33: Yes), the counter i = i + 1 (S34), the topic expansion is completed (S35), and the topic set expansion process (S2) is completed. In step S33, if i < N is not true (S33: No), the topic set expansion process (S2) is also completed.

[0063] As described above, the topic set expansion process (S2) is performed by the topic set expansion unit 30, and similar topics are added to the topics selected by the topic extraction unit 20, resulting in an expanded topic set.

[0064] Next, we will explain in detail the topic range comparison process (S4) performed by the topic range comparison unit 40.

[0065] Figure 10 is a flowchart showing the flow of the topic range comparison process by the topic range comparison unit.

[0066] In the topic range comparison process (S4), the topic range comparison unit 40 compares topic sets handled by multiple content items and detects topic ranges that clearly indicate the overlap of multiple topic sets.

[0067] Specifically, as shown in the flowchart of Figure 10, the topic range comparison unit 40 reads the topics of the N content items to be compared (S41), sets the counter i to 0 (S42), starts topic range comparison (S43), sets the counter j to j+1 (S44), starts reading the content items to be compared (S45), sets the counter k to 0 (S46), starts topic similarity comparison (S47), and reads the k-th topic T_ik of content i (S48). Then, if there is a topic T_jl in content j that is similar to topic T_ik (S49: Yes), (frequency of occurrence of topic T_ik on the platform) ≥ (frequency of occurrence of topic T_jl on the platform) (S50: Yes), and topic T_ik is included in the common topic set G_cij of content i and content j (S51: Yes), then topic T_ik is added to the common topic set G_cij (S52), and the process proceeds to step S55.

[0068] On the other hand, in step S50, if the frequency of occurrence of topic T_ik on the platform is not greater than or equal to the frequency of occurrence of topic T_jl on the platform (S50: No), if topic T_jl is included in the common topic set G_cij (S53: Yes), proceed to step S55; otherwise, add topic T_jl to the common topic set G_cij (S54) and proceed to step S55.

[0069] In step S55, if counter k < M (S55: Yes), counter k = k + 1 (S56), the topic similarity comparison is terminated (S57), and the process proceeds to step S58. Hereinafter, M is the number of topics in content i. On the other hand, if counter k < M is not the case in step S55 (S55: No), the process proceeds to step S58.

[0070] In step S58, topics from content j that were not included in the common topic set G_cij are added to topic set G_cijj (S58). Then, in step S59, topics from content i that were not included in the common topic set G_cij are added to topic set G_ciji (S59), and the process proceeds to step S60.

[0071] In step S60, if j < N (S60: Yes), the counter j = j + 1 (S61), the loading of the content to be compared is completed (S62), and the process proceeds to step S63. On the other hand, if j < N is not true in step S60 (S60: No), the process also proceeds to step S63.

[0072] In step S63, if i < (N-1) (S63: Yes), the counter i = i + 1 (S64), the comparison of topic ranges is completed (S65), and the topic range comparison process (S4) is terminated. Also, in step S63, if i < (N-1) is not true (S63: No), the topic range comparison process (S4) is terminated.

[0073] Note that the example operation shown in the flowchart of Figure 10 is an example where the number of content items used for comparison is two. However, the topic range comparison process (S4) can compare ranges between any number of content items, three or more. When comparing three or more N content items, this can be handled by considering the union of the topic sets of N-1 content items as one topic set, and comparing that topic set with the topic set of the remaining one content item.

[0074] For example, when comparing three content items A, B, and C, first compare content A and content B, and then find the union of topics T between content A and content B. Ａ ∪T Ｂ By creating this, we assume content A and B, and then we address the issue by comparing content A and B with content C.

[0075] In this way, the topic range comparison unit 40 performs the topic range comparison process (S4).

[0076] In this topic range comparison process (S4), the topic range comparison unit 40 can use only the topic set before expansion, only the topic set after expansion, or only the topic set after expansion for content that satisfies certain conditions. For example, this condition could be that the number of elements in the topic set before expansion is less than or equal to b (where b is any natural number).

[0077] Furthermore, the topic range comparison unit 40 can select common topics in the topic range comparison process (S4) as follows.

[0078] For example, the topic range comparison unit 40 can perform pattern matching and select topics with a perfect word match as common topics.

[0079] Figure 11 is a Venn diagram illustrating an example of selecting common topics based on exact word matches.

[0080] In the example shown in Figure 11, "DX" is the common topic.

[0081] Furthermore, the topic range comparison unit 40 can compare vectors for each word extracted as a topic, select the words that are closest to each other as similar topic pair candidates, and select two words whose distance from the similar topic pair candidates is less than a threshold as a common topic.

[0082] Furthermore, the topic range comparison unit 40 can determine common topics in the topic range comparison process (S4) as follows.

[0083] Figure 12 is an example of a Venn diagram used to determine common topics.

[0084] The topic range comparison unit 40 can determine that only topics common to the topic sets before expansion are common topics (Example 1). The topic range comparison unit 40 can also determine that only topics that were included in at least one content before expansion are common topics among the topic sets after expansion (Example 2). Furthermore, the topic range comparison unit 40 can also determine that all topics common to the topics after expansion are common topics (Example 3).

[0085] The topic range comparison unit 40 can detect relationships that could not be gleaned from the summary text alone by determining common topics using this expanded topic set. The topic range comparison unit 40 can also similarly determine topics specific to each piece of content.

[0086] Furthermore, the topic range comparison unit 40, in particular, by using the expanded topic set as described above in (Example 2) and (Example 3), makes it possible to examine the relationships between content even if the content contains only a small number of topics.

[0087] As described above, the relevance detection device to which the relevance detection method according to this embodiment is applied abstracts the target so that it can be applied to all content, regardless of the content's content, through the above-described action. As a result, especially when the content is educational content, it becomes possible to understand the overlap in the content of the educational content, so that learners can accurately determine which combination of courses they should take. For example, if 70% of the topics of Course A and Course B are the same, and the learner is not interested in the topics included only in Course B, then the learner can determine that they only need to take Course A.

[0088] Thus, with a relationship detection device to which the relationship detection method according to this embodiment is applied, the relationships between content become clear, allowing learners to understand the optimal combination of content and proceed with their learning efficiently.

[0089] The present invention is not limited to the embodiments described above, and in the implementation stage, the components can be modified and implemented without departing from the spirit of the invention. Furthermore, various inventions can be formed by appropriately combining the multiple components disclosed in the above embodiments. For example, some components may be deleted from all the components shown in the embodiments. Moreover, components from different embodiments may be appropriately combined.

[0090] 10 Relevance detection device 20 Topic extraction unit 30 Topic set expansion unit 40 Topic range comparison unit 100 Summary text, description text 102 Table of contents 104 Topics 108 Similar topics 110 Expanded topic set 114 Topic range 200 Computer 220 Input device 230 Memory 230 Output device

Claims

1. A relationship detection device comprising: a topic extraction unit that extracts topics dealt with by the content from documents attached to the content; a topic set expansion unit that adds similar topics to the extracted topics to obtain an expanded topic set; and a topic range comparison unit that compares the topic sets dealt with by multiple content items and detects topic ranges that clearly indicate the overlap of the multiple topic sets.

2. The association detection device according to claim 1, wherein the topic extraction unit extracts characteristic words from the document, compares the extracted words with the education-related document and other general documents, selects words that are more frequent in the education-related document from among the extracted words, creates a word vector based on the selected words, clusters similar words using the word vector, and extracts the word with the highest frequency of occurrence among the clustered words as the topic.

3. The relation detection device according to claim 1, wherein the topic set expansion unit vectorizes the topics of the content, compares the vectorized topics with word vectors in a pre-prepared topic dictionary, selects topics from the vectorized topics in order of similarity based on the results of the comparison, and adds the selected topics to the topics extracted by the topic extraction unit to obtain the expanded topic set.

4. A relationship detection method performed by a relationship detection device, wherein the processor of the relationship detection device extracts topics dealt with by the content from documents assigned to the content, adds similar topics to the extracted topics to obtain an expanded topic set, and compares multiple topic sets, each dealt with by multiple content, to detect topic ranges that clearly indicate the overlap of the multiple topic sets.

Images