Information processing device, information processing method, and program

Abstract

This enables the presentation of the level of abstraction of the words used to represent skills when creating a hierarchical skill list. [Solution] The skill word acquisition unit 2 acquires multiple skill words selected by the skill list creator from a list of multiple skill words. The skill description acquisition unit 3 acquires multiple skill descriptions generated using the generation AI 10. The distributed representation acquisition unit 4 acquires a distributed representation for each skill word, converted from each of the multiple skill descriptions using the embedding model 20. The abstraction level (standard deviation) calculation unit 5 calculates the standard deviation of the distributed representation in at least one dimension of the distributed representation (vector) of each of the multiple skill descriptions for each skill word. The standard deviation (magnitude) of at least one dimension corresponds to the abstraction level (height) of that skill word.

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] In a system for managing employees' skills, there is known a technique of performing syntactic analysis and semantic analysis on sentences written in a natural language such as a work history document, and storing words determined to be skills in a database as an individual's abilities (see, for example, Patent Document 1).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In order to accurately grasp employees' skills (abilities), it is necessary to manage skills in a hierarchical manner. When creating a hierarchical skill list, there is a task of creating a skill list with a hierarchical structure according to the abstraction level of skill words representing skills (an index indicating whether the word is a word with a broader meaning (upper layer) or a word with a specific and narrower meaning (lower layer)). However, if one is not familiar with the occupation, it is difficult to create a skill list with a hierarchical structure because the abstraction level (hierarchical structure) of skill words is unknown.

[0005] Therefore, an object of the present invention is to derive the abstraction level of words representing skills when creating a skill list with a hierarchical structure.

Means for Solving the Problems

[0006] The information processing device according to this invention is characterized by comprising: an acquisition unit that acquires a word and a distributed representation of a predetermined number or more descriptive sentences that describe the word; a derivation unit that derives a standard deviation from the distributed representation of each of the descriptive sentences in at least one dimension of the distributed representation of the predetermined number or more descriptive sentences; and a determination unit that associates the magnitude of the standard deviation derived by the derivation unit with the level of abstraction of the word corresponding to the descriptive sentence used when deriving the standard deviation. [Effects of the Invention]

[0007] According to this invention, it is possible to derive the level of abstraction of the words representing skills when creating a hierarchical skill list. [Brief explanation of the drawing]

[0008] [Figure 1] This is an external view showing the configuration of an information processing device according to an embodiment of the present invention. [Figure 2] This is a flowchart illustrating the operation of the information processing device according to this embodiment. [Figure 3] This is a schematic diagram showing an example of the initial screen according to this embodiment. [Figure 4] This is a schematic diagram showing an example of the display screen when selecting a skill word according to this embodiment. [Figure 5] This is a conceptual diagram illustrating the operation (explanatory text generation) according to this embodiment. [Figure 6] This is a conceptual diagram illustrating the operation (distributed representation) of the information processing device according to this embodiment. [Figure 7] This is a conceptual diagram showing the distribution of distributed representations according to this embodiment. [Figure 8] This is a conceptual diagram illustrating the method for calculating the level of abstraction index according to this embodiment. [Figure 9] This is a schematic diagram showing an example of a display screen for the determination result according to this embodiment. [Figure 10] This is a schematic diagram showing an example of a hierarchical skill list according to this embodiment. [Modes for carrying out the invention]

[0009] The information processing device in the present invention will be described with reference to Figure 1. The information processing device 1 is a computer equipped with a display device (not shown), an input device such as a keyboard (not shown), and an external storage device (not shown). As shown in Figure 1, the information processing device 1 includes a skill word acquisition unit (acquisition unit) 2, a skill description acquisition unit (acquisition unit) 3, a distributed representation acquisition unit (derivation unit) 4, an abstraction level (standard deviation) calculation unit (derivation unit) 5, a hierarchical determination unit (determination unit) 6, a determination result presentation unit (presentation unit) 7, and a storage unit 8. The skill word acquisition unit 2 acquires any multiple (at least two) skill words selected by the skill list creator from a skill word list 81 consisting of a list of multiple skill words representing various skills. The multiple skill words selected by the skill list creator are stored as a selected skill word list 82. The skill description acquisition unit 3 uses a generation AI (Artificial Intelligence) 10, such as ChatGPT (registered trademark), to acquire multiple sentences (skill descriptions) that explain skill words, generated from multiple selected skill word lists 82 acquired by the skill word acquisition unit 2. For example, 100 skill descriptions are acquired for a single skill word. Note that technologies such as web scraping may be used instead of the generation AI 10. Also, the number of skill descriptions acquired for each skill word may vary.

[0010] The distributed representation acquisition unit 4 obtains a distributed representation for each skill word from each of the multiple skill description sentences using an embedding model 20 such as text-embedding-ada-002. Typically, the distributed representation is a multidimensional vector (e.g., 1536 dimensions). It is known that the transformed distributed representation follows a normal distribution in each dimension. The abstraction level (standard deviation) calculation unit 5 calculates the standard deviation of the distributed representation (vector) in at least one dimension of the distributed representation (vector) of each of the multiple skill description sentences for each skill word. The magnitude of this standard deviation is taken as the abstraction level (height) of that skill word. Alternatively, for each skill word, the mean of the standard deviations of all dimensions in the distributed representation (vector) of each of the multiple skill description sentences may be calculated, and this mean of standard deviation may be taken as the abstraction level of that skill word. The hierarchy determination unit 6 determines, based on the skill abstraction level (height) of each skill word, whether the skill word belongs to a higher hierarchy (a word with a broader meaning) or a lower hierarchy (a word with a specific and narrower meaning). The judgment result presentation unit 7 presents the skill list creator with the judgment result indicating whether each skill word belongs to a higher or lower level.

[0011] Here, we will explain the level of abstraction of skill terms. The level of abstraction is an indicator that shows whether a skill term has a broader meaning (higher level) or a specific and narrower meaning (lower level). For example, the skill "sales" is a broader term (higher level), while the skills "domestic sales" and "international sales" are specific and narrower terms (lower level). In other words, the skill "sales" belongs to the higher level, and the skills "domestic sales" and "international sales" belong to the lower levels. By clarifying this hierarchical structure, it becomes possible to understand that an employee is not simply a "sales" employee, but rather, even if they are a "sales" employee, they specifically possess skills in "domestic sales" or "international sales." As mentioned above, it is desirable to create a hierarchical skill list in order to accurately understand the skills (abilities) of employees, but such a hierarchical structure differs depending on the job type. Since the person creating the skill list is not familiar with all job types, there is a risk that they will not be able to create this hierarchical skill list accurately. For example, in a skill list, it is desirable that the skills "Design and Development," "Construction Method Development," and "Design Supervision" be placed at a lower level than the skill "Design and Development," but there is a risk that the skill list creator may not be able to set this correctly. Therefore, in this embodiment, the level of abstraction of the skill word is derived, and the skill list creator is shown whether the skill word belongs to a higher or lower level, thereby supporting the creation of the hierarchical structure of the skill list.

[0012] The memory unit 8 includes a skill word list 81, a selected skill word list 82, a skill description 83, and distributed representation data 84. The memory unit 8 may be the computer's built-in memory or an external storage device. The skill word list 81 is a list of multiple pre-created skill words. The list of skill words may be arranged randomly, or they may be arranged according to some rule, such as alphabetical order. The skill list creator selects at least two or more skill words from the skill word list 81 to determine whether they belong to a higher or lower level.

[0013] The selected skill word list 82 is a list of skill words selected by the skill list creator. The skill description 83 is a skill description generated from the selected skill word list 82 obtained by the skill word acquisition unit 2 using a generation AI 10 such as ChatGPT (registered trademark). The distributed representation data 84 is a distributed representation (vector) converted from multiple skill descriptions for each skill word using the embedding model 20.

[0014] The information processing device 1 performs operations according to the flowchart shown in Figure 2. The information processing device 1 displays a display screen 30 as shown in Figure 3 on a display device (not shown) (step S10). On the display screen 30, as shown in Figure 3, there is a window W1 that displays a list of skill words, a window W2 that displays a list of selected skill words, a window W3 that displays the judgment result, and a group of buttons B for instructing various operations. Window W1 displays a list of pre-created skill words read from the skill word list 81.

[0015] Next, the skill word acquisition unit 2 acquires any multiple (at least two) skill words selected by the skill list creator from the list of multiple skill words displayed in window W1 (step S12). The skill list creator selects at least two skill words for which they want to determine the higher and lower levels. Figure 4 shows the state in which the skill list creator has selected three skill words, "design and development," "construction method development," and "design supervision," from window W1. Note that in the example shown in Figure 4, adjacent skill words are selected, but the system is not limited to this, and non-adjacent skill words may also be selected. The skill word acquisition unit 2 displays the selected skill words in window W2 and stores them in the storage unit 8 as the selected skill word list 82 (step S14).

[0016] Next, the skill description acquisition unit 3 determines whether the "Analysis Execution" button B1 in the button group B shown in FIG. 4 has been pressed (step S16). When the "Analysis Execution" button B1 is pressed by the skill list creator (YES in step S16), the skill description acquisition unit 3 gives the selection skill word list 82 including the selected multiple skill words to the generation AI 10, and acquires a plurality of sentences (skill descriptions; for example, 100 sentences) that explain each skill word generated by the generation AI 10 (step S18). For example, as shown in FIG. 5, an instruction such as "Please generate n sentences (for example, n = 100) that explain 'Programming' as a skill" is given to the generation AI 10 such as ChatGPT. The generation AI generates n skill descriptions such as "Make the computer execute a specific task...", "For creating software and websites...", "Utilize problem-solving and communication technologies..." as descriptions of "Programming" according to the instruction. When "C language", "Python", etc. are selected in addition to "Programming", n skill descriptions are also generated for "C language" and "Python" respectively. The skill description acquisition unit 3 stores the skill descriptions from the generation AI 10 in the storage unit 8 as skill descriptions 83.

[0017] Next, the distributed representation acquisition unit 4 gives each selected skill word a plurality of respective skill descriptions and provides them to the embedding model 20, and acquires the distributed representations for the plurality of respective skill descriptions converted by the embedding model 20 (step S20). For example, as shown in FIG. 6, when giving the embedding model 20 n skill descriptions such as "To make a computer execute a specific task...", "For creating software and websites...", "Utilizing problem-solving and communication technologies..." as the description of "Programming", the embedding model 20 converts them into distributed representations (vectors) such as [0.749, -1.383,..., -1.029, 0.601], [-0.153, 2.858,..., 1.714, -0.180], [2.637, -0.768,..., -0.826, -1.849]. The distributed representation acquisition unit 4 stores the distributed representations from the embedding model 20 in the storage unit 8 as distributed representation data 84. <{

[0018] Next, the abstraction level (standard deviation) calculation unit 5 calculates the standard deviation of the distributed representation (vector) in at least one dimension in the distributed representations (vectors) of the plurality of respective skill descriptions for each selected skill word (step S22). FIG. 7 shows, as an example, the distribution of 10 dimensions among the distributed representations (vectors) of 100 skill descriptions generated for the skill word "Programming". It can be seen that the distribution of the distributed representation (vector) in any dimension is close to a normal distribution. In this embodiment, as shown in FIG. 8, in order to estimate the abstraction level of the skill word from the degree of variation in the distribution of the distributed representation (vector), the standard deviation σ is calculated from the distribution of the distributed representation (vector) in at least one dimension. In the case of an abstract skill word (for example, Programming), there is also variation in the content of the skill description that is the data source of the distributed representation, and the standard deviation σ1 in each dimension becomes large. On the other hand, in the case of a more specific skill word (for example, C language), the variation in the content of the skill description also becomes small, and the standard deviation σ2 in each dimension becomes small (σ1 > σ2).

[0019] Next, the hierarchy determination unit 6 compares the skill abstraction level (standard deviation) of each skill word and determines whether the skill word belongs to a higher hierarchy (a word with a broader meaning) or a lower hierarchy (a word with a specific and narrower meaning) (step S24). In this embodiment, the magnitude of the standard deviation σ of at least one dimension is associated with the level of abstraction of that skill word. More specifically, a larger standard deviation σ corresponds to a higher level of abstraction for that skill word, and a smaller standard deviation σ corresponds to a lower level of abstraction. In this embodiment, a predetermined threshold is set, and if the skill abstraction level is greater than or equal to the predetermined threshold, it is determined to be a higher hierarchy; if it is less than the predetermined threshold, it is determined to be a lower hierarchy. The predetermined threshold can be determined by methods such as (maximum value + minimum value) / 2 of the abstraction level of each skill word, or by using the k-means method or Otsu's binarization method, as long as the maximum value of the abstraction level of each skill word ≥ the predetermined threshold ≥ the minimum value. Alternatively, if there is even a slight difference in the level of abstraction, skill words with a higher level of abstraction may be determined as higher-level, and skill words with a lower level of abstraction may be determined as lower-level. Not all skill words will be in the higher (or lower) level; at least one of each should be classified as either higher-level or lower-level. Finally, the judgment result presentation unit 7 presents the skill list creator with the judgment result of which skill words are in the higher level or lower level (step S26). Figure 9 shows an example of the display of the judgment result (analysis result) when "design and development," "construction method development," and "design supervision" are selected as skill words. The judgment result presentation unit 7 displays in window W3 on the display screen 30 that the skill word "design and development" has been determined to be in the higher level, and the skill words "construction method development" and "design supervision" have been determined to be in the lower level. Additionally, although not shown in the diagram, if "Sales," "Domestic Sales," and "International Sales" are selected as skill words, the judgment result will be displayed in window W3 on display screen 30, indicating that the skill word "Sales" was determined to be a higher level, and the skill words "Domestic Sales" and "International Sales" were determined to be lower levels.Although not shown in the diagram, if "programming," "C language," and "Python" are selected as skill words, the result of the determination will be displayed in window W3 on display screen 30, indicating that the skill word "programming" was determined to be a higher level, and the skill words "C language" and "Python" were determined to be lower levels.

[0020] Figure 10 shows an example of a hierarchical skill list created by a skill list creator based on the level of abstraction of the skill words presented by the information processing device 1 according to this embodiment. For example, a hierarchical skill list is created by referring to the judgment result (analysis result) displayed in window W3 as shown in Figure 9. As shown in Figure 10, the skill list creator creates a hierarchical skill list in which the skill word "design and development" is set as the upper level and the skill words "construction method development" and "design supervision" are set as lower levels. Also, as mentioned above, if "sales," "domestic sales," and "overseas sales" are selected as skill words, the information processing device 1 determines that the skill word "sales" is the upper level and the skill words "domestic sales" and "overseas sales" are lower levels. Therefore, as shown in Figure 10, the skill list creator creates a hierarchical skill list in which the skill word "sales" is set as the upper level and the skill words "domestic sales" and "overseas sales" are set as lower levels. Furthermore, as mentioned above, when the skill words "programming," "C language," and "Python" are selected, the information processing device 1 determines that the skill word "programming" is in the higher hierarchy and the skill words "C language" and "Python" are in the lower hierarchy. Therefore, the skill list creator creates a skill list with a hierarchical structure in which the skill word "programming" is set in the higher hierarchy and the skill words "C language" and "Python" are set in the lower hierarchy, as shown in Figure 10.

[0021] According to the embodiment described above, multiple skill descriptions are obtained for multiple skill words, and distributed representations are obtained for each of the multiple skill descriptions. A standard deviation is derived from the distributed representation of each skill description in at least one dimension of the distributed representations of the multiple skill descriptions. The magnitude of the derived standard deviation is associated with the level of abstraction of the skill word corresponding to the skill description used to derive the standard deviation. This makes it possible to derive the level of abstraction of skill words when creating a hierarchical skill list. Therefore, it is possible to show the skill list creator whether a skill word is in a higher hierarchical level (a word with a broader meaning) or a lower hierarchical level (a word with a specific and narrower meaning) according to its level of abstraction, thereby supporting the creation of a hierarchical skill list.

[0022] In the embodiment described above, it is also possible to identify (distinguish) which industry (category) a skill word belongs to, for example, construction, real estate, medical / pharmaceutical, production / logistics, etc., and to present the industry (category) of the selected skill word. By identifying and presenting the industry, it becomes possible to categorize skill words within the same industry into higher and lower levels.

[0023] Furthermore, in the embodiment described above, the skill word list 81 is created in advance, and subsequent processing is performed when the "Analysis Execution Button" B1 is pressed. However, the system is not limited to this, and it is also possible to perform the acquisition and saving of skill descriptions 83 for all skill words in advance, or the acquisition and saving of distributed representation data 84 for all skill descriptions 83 in advance, or the calculation of the level of abstraction for all distributed representation data 84 in advance. This makes it possible to shorten the time required from when the skill list creator selects a skill word until the information processing device 1 presents the judgment result (analysis result).

[0024] Furthermore, while the above-described embodiment explained skill words representing employee skills (abilities), it is not limited to these. As long as the object of analysis has a hierarchical structure, it may also be words representing geographical hierarchies such as "Eurasian continent," "Europe," "Germany," and "Munich," or words representing hierarchical structures of company organizations such as "company name," "branch name," and "department name," or words representing classification hierarchies (kingdom, phylum, class, order, family, genus, species) of animals and plants.

[0025] Furthermore, in the above-described embodiment, the skill list creator manually created the skill list based on the hierarchical structure derived by the information processing device 1. However, the system is not limited to this, and the information processing device 1 may create and output a hierarchical skill list based on the judgment result (analysis result).

[0026] Furthermore, in the embodiment described above, a series of processes are performed on the information processing device 1 (computer) operated by the user, who is the skill list creator. However, for example, the user's computer may access a server on the network via a general web browser, the server may perform the series of processes described above, and send the judgment result (analysis result) back to the user's web browser. In other words, the processes performed by the information processing device 1 may be distributed and executed between the user's computer and the server on the network.

[0027] Furthermore, it goes without saying that the detailed configuration and operation of each component of the information processing device 1 in the above embodiments can be appropriately modified without departing from the spirit of the present invention. Although embodiments of the present invention have been described, the scope of the present invention is not limited to the embodiments described above, but includes the scope of the invention as described in the claims and its equivalents. [Explanation of Symbols]

[0028] 1... Information processing device, 2... Skill word acquisition unit (acquisition unit), 3... Skill description acquisition unit (acquisition unit), 4... Distributed representation acquisition unit (derivation unit), 5... Abstraction level (standard deviation) calculation unit (derivation unit), 6... Hierarchy determination unit (determination unit), 7... Determination result presentation unit (presentation unit)

Claims

1. An acquisition unit that acquires a word and a predetermined number of or more descriptive sentences that explain the word, A derivation unit that derives the standard deviation from the distributed representation of each of the predetermined number of explanatory statements in at least one dimension of the distributed representation of the explanatory statements, A determination unit that associates the magnitude of the standard deviation derived by the derivation unit with the level of abstraction of the word corresponding to the explanatory text used when deriving the standard deviation, Equipped with an information processing device.

2. The determination unit, The system is configured such that a high level of abstraction corresponds to a large standard deviation, and a low level of abstraction corresponds to a small standard deviation. The information processing apparatus according to feature 1.

3. The aforementioned derivation section is, When using standard deviations of multiple dimensions, the mean of the standard deviations of the multiple dimensions is derived. The determination unit, The magnitude of the average value of the standard deviation derived by the derivation unit is associated with the level of abstraction of the words corresponding to the explanatory text used when deriving the average value of the standard deviation. The information processing apparatus according to feature 1.

4. The acquisition unit is, Obtaining distributed representations of multiple words and a predetermined number or more descriptive sentences that explain each of the multiple words. The information processing apparatus according to feature 1.

5. The aforementioned words are skill-related terms concerning employee skills. The information processing apparatus according to feature 1.

6. The aforementioned skill words are, It has a hierarchical structure in terms of its narrow or broad definition. The determination unit, The abstraction level of the skill words derived by the derivation unit is associated with the hierarchical structure. The information processing apparatus according to feature 5.

7. The system further includes a presentation unit that outputs the associated skill words and the hierarchical structure. The information processing apparatus according to feature 6.

8. The explanation for the aforementioned word is: Generated by generative AI (Artificial Intelligence), The information processing apparatus according to feature 1.

9. An information processing method by the control unit of an information processing device, A step of obtaining a word and a predetermined number of or more explanatory sentences that describe the word, A step of deriving the standard deviation from the distributed representation of each of the predetermined number of explanatory statements in at least one dimension of the distributed representation of the explanatory statements, The steps include: associating the magnitude of the derived standard deviation with the level of abstraction of the word corresponding to the explanatory text used in deriving the standard deviation; An information processing method characterized by including

10. In the computer of a device equipped with a control unit, A function for obtaining a word and a predetermined number of descriptive sentences that explain the word, each containing a distributed representation. A derivation function that derives the standard deviation from the distributed representation of each of the predetermined number of descriptive sentences in at least one dimension of the distributed representation of the descriptive sentences, A determination function that associates the magnitude of the standard deviation derived by the derivation function with the level of abstraction of the word corresponding to the explanatory text used in deriving the standard deviation. A program characterized by its ability to achieve this.

Images