Information processing method and information processing system
The information processing method evaluates the uniqueness of answers by comparing theme and answer vectors, using low originality answers as references to prevent misclassification and enhance creativity through lighting adjustments, addressing the inaccuracies in existing linguistic creativity evaluation methods.
Patent Information
- Application Number
- PCT/JP2024/016514
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-05-23
- Filing Date
- 2024-04-26
- Publication Date
- 2024-11-28
Smart Images

Figure JP2024016514_28112024_PF_FP_ABST
Abstract
Description
Information processing method and information processing system
[0001] The present disclosure relates to techniques for evaluating people's ideas.
[0002] In recent years, various studies have been conducted to evaluate people's linguistic creativity. For example, Non-Patent Document 1 discloses a technology that addresses the subjective limitations in evaluating people's creativity by utilizing recent research on automatically scoring linguistic creativity using semantic distance.
[0003] Automating creativity assessment with SemDis: An open platform for computing semantic distance, Roger E. Beaty & Dan R. Johnson, Published online: 31 August 2020
[0004] Creativity can be evaluated from three main perspectives: originality, fluency, and flexibility. However, distance in a semantic space, such as semantic distance, is primarily related to originality. Therefore, strictly speaking, creativity as referred to in Non-Patent Document 1 is considered to be closely related to originality.
[0005] Even if an answer created based on a certain topic has commonplace content, it is undesirable to judge the answer to be highly unique simply because the answer's vector distance from the topic is large. The technology in Non-Patent Document 1 does not take such issues into consideration, and therefore has the problem of being unable to accurately judge the uniqueness of an answer created from a certain topic.
[0006] The present disclosure provides techniques for more accurately assessing the originality of answers created from a given subject.
[0007] An information processing method according to one aspect of the present disclosure is a computer-implemented information processing method, which includes obtaining a first vector corresponding to a first answer text indicating an answer to a topic, obtaining a second vector corresponding to a second answer text indicating an answer associated with the topic, and evaluating the uniqueness of the answer indicated by the first answer text based on the magnitude of the similarity between the first vector and the second vector. Note that this comprehensive or specific aspect may be realized as an apparatus, a system, an integrated circuit, a computer program, or a computer-readable recording medium, or may be realized as any combination of a method, an apparatus, a system, an integrated circuit, a computer program, and a recording medium. The computer-readable recording medium may include a non-volatile recording medium such as a CD-ROM (Compact Disc-Read Only Memory).
[0008] The present disclosure allows for a more accurate assessment of the originality of answers created from a given subject.
[0009] 1 is a diagram illustrating an example of a network configuration of an information processing system according to an embodiment of the present disclosure. FIG. 2 is a block diagram illustrating an example of a detailed configuration of an information processing system. FIG. 3 is a diagram illustrating an example of a data configuration of a reference text table. FIG. 4 is a diagram illustrating an example of a data configuration of an evaluation result table. FIG. 5 is a sequence diagram illustrating a processing flow of the information processing system. FIG. 6 is a flowchart illustrating an example of a processing of the information processing system when acquiring a reference answer. FIG. 7 is a flowchart illustrating details of the processing of step S205 shown in FIG. 6 and step S405 shown in FIG. 7. FIG. 8 is an explanatory diagram of the processing of FIG. 8. FIG. 9 is a flowchart illustrating details of the processing of step S407 in FIG. 7. FIG. 10 is an explanatory diagram of the processing of FIG. 10. A flowchart illustrating an example of a lighting control processing. FIG. 11 is a diagram illustrating an example of a display screen displayed on a user's terminal in step S409 in FIG. 7. FIG. 12 is a diagram illustrating an example of a booth.
[0010] (Findings underlying the present disclosure) The present inventors are conducting research into having a computer evaluate the degree of originality of a user's response when a certain topic is given to the user. When a computer performs such an evaluation, the following method is envisioned. The computer represents each of the topic and the response to that topic as a word vector. Next, the computer calculates the cosine distance between both word vectors. The larger the cosine distance, the more original the response is determined to be.
[0011] A specific example is given below. Suppose the topic "How to use a futon" is given to a user and the answer "lay it on the floor" is obtained. The answer "lay it on the floor" to this topic is considered commonplace by many people and is recognized as having low uniqueness. However, if "How to use a futon" and "lay it on the floor" are each represented as word vectors, the cosine distance between the two word vectors may be far apart. In this case, the answer "lay it on the floor" will be determined to be highly unique. This makes it impossible to accurately evaluate the uniqueness of answers to the topic.
[0012] Therefore, the inventors discovered that, rather than using a certain topic as a standard, answers with low originality for that topic can be collected in advance, and answers created by users can be evaluated based on the collected answers with low originality, which led to the present disclosure.
[0013] (1) An information processing method in one aspect of the present disclosure is an information processing method in a computer, which obtains a first vector corresponding to a first answer text indicating an answer to a topic, obtains a second vector corresponding to a second answer text indicating an answer associated with the topic, and evaluates the uniqueness of the answer indicated by the first answer text based on the degree of similarity between the first vector and the second vector.
[0014] According to this configuration, the uniqueness of an answer is evaluated using the vector similarity between the first answer text and a second answer text associated with the topic, which indicates an answer considered to be less unique, rather than the vector similarity between the topic and a first answer text indicating an answer to the topic. This reduces the risk that an answer considered to be less unique is erroneously evaluated as being highly unique. As a result, the uniqueness of an answer created from a certain topic can be more accurately evaluated.
[0015] (2) In the information processing method described in (1) above, the second answer text may include a plurality of texts, and the second vector may be a vector corresponding to a center of gravity position determined from a plurality of vectors generated by converting each of the plurality of texts into a vector.
[0016] According to this configuration, the second vector is composed of one vector that represents the characteristics of the second answer text, which makes it easy to calculate the similarity between the first vector and the second vector.
[0017] (3) In the information processing method described in (1) or (2) above, the similarity may be a cosine similarity derived based on the first vector and the second vector.
[0018] According to this configuration, the similarity between the first vector and the second vector can be accurately quantified, and the uniqueness of the first answer text can be evaluated more accurately.
[0019] (4) In the information processing method described in any one of (1) to (3) above, the first answer text may be obtained by converting voice data based on a user's speech into text data.
[0020] According to this configuration, the first answer text can be input by voice, which reduces the burden of inputting the first answer text.
[0021] (5) In the information processing method described in any one of (1) to (4) above, evaluating the uniqueness of the answer may include deriving a score according to the degree of uniqueness of the answer.
[0022] According to this configuration, the first answer text is evaluated using a score according to the degree of uniqueness, which makes it easy to evaluate the uniqueness of the first answer text.
[0023] (6) In the information processing method described in (5) above, when a first condition indicating that the score is low is met, a lighting device that illuminates the vicinity of the user may be controlled to lower the color temperature of the illumination light and increase the illuminance.
[0024] According to this configuration, when the score is low, the color temperature of the illumination light is lowered and the lighting device is controlled to increase the illuminance, thereby enhancing the user's creativity and encouraging the creation of a highly original first answer text.
[0025] (7) In the information processing method described in (6) above, it may be determined that the first condition is satisfied when the score is smaller than a threshold value.
[0026] According to this configuration, it is possible to accurately determine whether the score is low or not.
[0027] (8) In the information processing method described in (6) above, obtaining the first vector may include obtaining a plurality of first answer texts, evaluating the uniqueness of the answers may include deriving a plurality of scores for a plurality of answers corresponding to each of the plurality of first answer texts obtained during a certain period of time, and determining whether the first condition is met may include determining that the first condition is met if, during the certain period of time, a predetermined number or more of the plurality of scores have a score below a threshold.
[0028] According to this configuration, the lighting device is controlled based on the evaluation results of a plurality of answers created over a certain period of time, thereby avoiding situations in which the lighting device control is changed frequently.
[0029] (9) In the information processing method described in any one of (1) to (8) above, the first vector may be the vector that has the lowest similarity to the second vector among a plurality of vectors corresponding to each of a plurality of words included in the first answer text.
[0030] According to this configuration, the characteristics of the first answer text can be expressed by one index, so even if the first answer text is a long sentence, it is easy to calculate the similarity while capturing the characteristics of the long sentence.
[0031] (10) In the information processing method according to any one of (1) to (9) above, a display screen for displaying an evaluation result of the originality of the answer may be displayed on the display.
[0032] According to this configuration, the evaluation result for the answer can be fed back to the user.
[0033] (11) In the information processing method described in any one of (1) to (10) above, the first vector may include a plurality of word vectors calculated for each of a plurality of words included in the first answer text for each of a plurality of corpora, and the second vector may include a plurality of reference points corresponding to the plurality of corpora, and the evaluation of the uniqueness may include calculating a plurality of evaluation values related to uniqueness corresponding to the plurality of corpora by comparing the plurality of word vectors with the plurality of reference points for each of the plurality of corpora, and calculating a final evaluation value from the plurality of evaluation values.
[0034] According to this configuration, the final evaluation value is calculated from evaluation values corresponding to a plurality of corpora, so that variations due to corpora are suppressed and the uniqueness of the answer can be calculated with high accuracy.
[0035] (12) In another aspect of the present disclosure, an information processing system is an information processing system having a processor, wherein the processor performs a process of acquiring a first vector corresponding to a first answer text indicating an answer to a topic, acquiring a second vector corresponding to a second answer text indicating an answer associated with the topic, and evaluating the uniqueness of the answer indicated by the first answer text based on the degree of similarity between the first vector and the second vector.
[0036] This configuration makes it possible to provide an information processing system that can more accurately evaluate the originality of answers created from a certain theme.
[0037] The present disclosure can also be realized as an information processing program that causes a computer to execute each of the characteristic components included in such an information processing system, or as an information processing device that operates by this information processing program. Needless to say, such a computer program can be distributed on a computer-readable non-transitory recording medium such as a CD-ROM or via a communication network such as the Internet.
[0038] Note that each of the embodiments described below represents a specific example of the present disclosure. The numerical values, shapes, components, steps, and step orders shown in the following embodiments are merely examples and are not intended to limit the present disclosure. Furthermore, among the components in the following embodiments, components that are not described in the independent claims that represent the highest concept are described as optional components. Furthermore, in all of the embodiments, the respective contents can be combined.
[0039] (Embodiment) FIG. 1 is a diagram illustrating an example of a network configuration of an information processing system 1 according to an embodiment of the present disclosure. The information processing system 1 is a system that provides a user with a topic and evaluates the user's response to that topic. The information processing system 1 includes a terminal 100, a reference calculation server 200, a storage unit 300, an originality evaluation server 400, a lighting control server 500, and a lighting device 600. The terminal 100 to the lighting device 600 are connected to each other so as to be able to communicate with each other via a network. The network is, for example, a wide-area communication network including the Internet and a mobile phone communication network. In FIG. 1, the terminal 100 and the lighting device 600 are installed in a space where a user is located. An example of the space is a booth 1500, which will be described later in FIG. 14.
[0040] The terminal 100 is a computer used by a user. The terminal 100 may be, for example, a portable computer such as a smartphone or tablet computer, or may be a desktop computer. The terminal 100 includes a microphone, a keyboard, a USB memory, a monitor, and the like. The microphone acquires voice data spoken by the user as an answer to the topic. The keyboard acquires the text of the answer entered by the user as an answer to the topic. Hereinafter, this text will be referred to as target text. The target text is an example of first text. The USB memory stores the target text. The monitor displays a display screen showing the evaluation results for the user's answer.
[0041] The reference calculation server 200, the originality evaluation server 400, and the lighting control server 500 are configured as computers such as cloud servers. The storage unit 300 is configured as a computer equipped with a rewritable non-volatile storage device such as a hard disk drive (HDD) or a solid state drive. The reference calculation server 200 calculates a reference score, which will be described later. The originality evaluation server 400 evaluates the originality of the user's answer. The lighting control server 500 controls the lighting device 600 based on the originality evaluation result. The storage unit 300 stores data required to evaluate the originality of the answer.
[0042] The lighting device 600 illuminates the inside of the booth 1500 shown in FIG. 14 under the control of the lighting control server 500 .
[0043] FIG. 2 is a block diagram showing an example of a detailed configuration of the information processing system 1.
[0044] The terminal 100 includes an input unit 101 , a communication unit 102 , a screen generation unit 103 , a screen control unit 104 , and a display unit 105 .
[0045] The input unit 101 is composed of a microphone, keyboard, USB memory, etc. as shown in Figure 1. The input unit 101 acquires target text. The input unit 101 acquires topic data indicating a topic input by a user. The topic data may be text or audio data. The communication unit 102 is composed of a communication circuit that connects the terminal 100 to a network. The screen generation unit 103 generates a display screen that shows the evaluation results of the originality of the user's answer. The screen control unit 104 displays the display screen generated by the screen generation unit 103 on the display unit 105. The display unit 105 is composed of the monitor shown in Figure 1, and displays the display screen under the control of the screen control unit 104.
[0046] The reference calculation server 200 includes a reference calculation unit 201 , a communication unit 202 , a data conversion unit 203 , and a theme setting unit 204 .
[0047] The reference calculation unit 201 calculates a reference point, which is a representative value of a reference vector corresponding to the theme data set by the theme setting unit 204. The reference point is a point in vector space corresponding to a reference answer text indicating a reference answer, which is an answer that is associated with the theme and has low uniqueness. The reference answer text is an example of a second answer text. The reference point is an example of a second vector. The reference vector is calculated using a method for vectorizing words, such as Word2vec.
[0048] The communication unit 202 is configured with a communication circuit that connects the reference calculation server 200 to a network. The communication unit 202 receives subject data transmitted from the terminal 100. The communication unit 202 receives voice data of a reference answer or reference text transmitted from the terminal 100.
[0049] When the subject data received by the communication unit 202 is voice data, the data conversion unit 203 converts the voice data into text data. The data conversion unit 203 obtains the reference text by converting the voice data of the reference answer received by the communication unit 202 into text data.
[0050] The subject setting unit 204 sets the subject data received by the communication unit 202. The criteria calculation unit 201, the data conversion unit 203, and the subject setting unit 204 may be realized by a processor such as a CPU executing an information processing program, or may be configured by a dedicated hardware circuit.
[0051] The originality evaluation server 400 includes an originality evaluation unit 401 , a communication unit 402 , a data conversion unit 403 , and a theme setting unit 404 .
[0052] The uniqueness evaluation unit 401 calculates an evaluation value for evaluating the uniqueness of the answer indicated by the target text based on the magnitude of similarity between the target vector generated by the data conversion unit 403 and the reference vector calculated by the reference calculation server 200. For example, the similarity is defined as cosine similarity. If the target vector is a and the reference vector is b, the cosine similarity is expressed as a·b / |a|·|b|. The larger the cosine similarity, the greater the similarity between the target vector and the reference vector, and therefore the lower the uniqueness of the answer. Therefore, the evaluation value is defined so that the smaller the cosine similarity, the larger the value. For example, the evaluation value is defined as (1 - cosine similarity) x 50.
[0053] The communication unit 402 is configured with a communication circuit that connects the originality evaluation server 400 to a network, and receives the reference score calculated by the reference calculation server 200 from the storage unit 300. The communication unit 402 receives the subject data transmitted from the terminal 100. The communication unit 402 receives the answer voice data or target text transmitted from the terminal 100.
[0054] When answer voice data is transmitted from the terminal 100, the data conversion unit 403 acquires the target text by converting the voice data into text data. The data conversion unit 403 calculates a target vector that represents the target text as a vector. The target vector is vectorized using a vectorization method such as Word2vec. The target vector is an example of a first vector.
[0055] The subject setting unit 404 sets the subject data received by the communication unit 402 .
[0056] The uniqueness evaluation unit 401, the data conversion unit 403, and the theme setting unit 404 may be realized by a processor such as a CPU executing an information processing program, or may be realized by a dedicated hardware circuit.
[0057] The lighting control server 500 includes a lighting control unit 501 and a communication unit 502. The lighting control unit 501 generates a control command for setting the illuminance and color temperature of the lighting device 600 based on the evaluation value calculated by the originality evaluation server 400. The communication unit 502 transmits the control command generated by the lighting control unit 501 to the lighting device 600. The communication unit 502 receives the evaluation value calculated by the originality evaluation server 400 from the storage unit 300. The lighting control unit 501 may be realized by a processor such as a CPU executing an information processing program, or may be configured as a dedicated hardware circuit. The communication unit 502 is configured as a communication circuit and transmits the control command to the lighting device 600.
[0058] The storage unit 300 includes a storage unit 301 and a communication unit 302. The storage unit 301 is configured as a rewritable non-volatile storage device. The storage unit 301 stores a reference text table T1 shown in FIG. 3 and an evaluation result table T2 shown in FIG. 4. The storage unit 301 stores a learning model used by the reference calculation server 200 and the uniqueness evaluation server 400 when calculating reference vectors and target vectors. For example, the learning model may be a learning model created in advance using machine learning technology to calculate the vector of an input word from the text of that word. In this embodiment, a model that represents words as vectors using Word2vec is used as the learning model.
[0059] 3 is a diagram showing an example of the data configuration of the reference text table T1. The reference text table T1 stores reference texts that indicate answers with low originality for each of a plurality of topics.
[0060] The reference text table T1 stores themes, reference texts, and elements of reference vectors with dimensions 1 to 300, in association with each other. For example, for the theme "pot," "boil vegetables" and "boil water" are stored as reference texts. In this way, the reference texts indicate answers with low originality that users can easily associate with the theme. Here, the reference vector is composed of 300-dimensional elements, but this is just an example, and the reference vector may be composed of elements with any number of dimensions other than 300.
[0061] FIG. 4 is a diagram showing an example of the data configuration of the evaluation result table T2. The evaluation result table T2 stores target texts indicating user answers for each of a plurality of topics, in association with evaluation values for those answers. The evaluation result table T2 includes fields for topics, target texts, evaluation words, evaluation values, and users. The evaluation words are words extracted from the target text for evaluation. The user field stores the user's identification information (e.g., name). For example, the record showing the evaluation result in the first row records that the user "Bob" answered "to make it the head of the robot" for the topic "pot," that "robot" and "head" were extracted as evaluation words, and that an evaluation value of "75" was calculated.
[0062] The user who answered can be estimated from the identification information assigned to the terminal 100 of that user. For example, if the uniqueness evaluation unit 401 obtains the answer "Make it the head of a robot" from Bob's terminal 100, it can estimate that the user who answered is Bob. When this aspect is adopted, the storage unit 300 can store user information that associates the identification information of the terminal 100 with the identification information of the user. Then, the uniqueness evaluation unit 401 can identify the identification information of the user that corresponds to the identification information of the terminal 100 from the user information.
[0063] The user who has answered may also be estimated by voiceprint matching. For example, if the storage unit 300 stores voiceprint information for each user, the uniqueness evaluation unit 401 may estimate the user by matching the voice data of the answer transmitted from the terminal 100 with the voiceprint information. This mode is applied when a single terminal 100 acquires voice data of answers from multiple users, rather than when each user inputs an answer using their own terminal 100.
[0064] 5 is a sequence diagram showing the processing flow of the information processing system 1. First, in step S201, the terminal 100 is started up. After starting up, the user decides what kind of theme to use and inputs theme data indicating the decided theme into the terminal 100. Here, audio data of the theme is input as the theme data. The input theme data is transmitted to the reference calculation server 200 (step S202). Note that the theme data may be text data.
[0065] Next, the user inputs voice data of reference answers, which are answers with low originality that are associated with the determined topic, into the terminal 100. Here, the user inputs as many reference answers as he or she can think of. The voice data of the input reference answers is transmitted to the reference calculation server 200 (step S203). The processing of steps S202 to S208 will be described in detail below with reference to FIG. 6.
[0066] 6 is a flowchart showing an example of the processing of the information processing system 1 when acquiring a reference answer. First, in step S202, the subject setting unit 204 of the reference calculation server 200 acquires the subject data transmitted from the terminal 100 using the communication unit 202, and sets the subject indicated by the acquired subject data as the target subject.
[0067] Next, in step S203, the data conversion unit 203 of the reference calculation server 200 acquires the voice data of the reference answer transmitted from the terminal 100 using the communication unit 202. Here, an example is shown in which the voice data of the reference answer is acquired from the terminal 100 of the user who will provide the answer that will be the subject of originality evaluation. However, this is just one example, and the voice data of the reference answer may be acquired from the terminals 100 of an unspecified number of general users who are different from the user. For example, the reference answer may be an answer collected through crowdsourcing.
[0068] Next, in step S204, the data conversion unit 203 performs processing to convert the voice data of the reference answer into text data, thereby obtaining the reference text.
[0069] In this example, audio data of the reference answer is acquired, but this is just an example, and text data of the reference answer may be acquired. In this case, instead of the processes of steps S203 and S204, a process of acquiring the reference text is executed.
[0070] Next, in step S205, the data conversion unit 203 converts the reference text into a reference vector by inputting the reference text into the learning model stored in the storage unit 300. Details of this process will be described later with reference to FIG.
[0071] Next, in step S206, the reference calculation unit 201 determines whether all reference texts indicating all reference answers acquired in step S203 have been converted into reference vectors. If all reference texts have not been converted into reference vectors (NO in step S206), the process returns to step S205, where the next reference text is converted into a reference vector. If all reference texts have been converted into reference vectors (YES in step S206), the reference calculation unit 201 calculates a reference point using all the reference vectors. The reference point is, for example, the center of gravity of all the reference vectors in vector space.
[0072] Next, in step S208, the reference calculation unit 201 stores the reference points in the storage unit 300. At this time, the subject data acquired in step S202, the reference text acquired in step S203, and the reference vector converted in step S205 are stored in the reference text table T1. As a result, the reference texts are accumulated in the reference text table T1.
[0073] Next, the process of evaluating the answer will be described. The process from step S401 onwards shown in Fig. 5 shows the process of evaluating the answer. The processes in steps S401 and S402 are the same as steps S201 and S202.
[0074] In step S403, the user inputs voice data of the answer to the task indicated by the task data input in step S402 into the terminal 100. Note that target text may be input instead of the voice data of the answer. Details of the processing of steps S402 to S409 will be explained below with reference to FIG. 7.
[0075] 7 is a flowchart showing an example of the process of the information processing system 1 when acquiring an answer. In step S403, the data conversion unit 403 of the originality evaluation server 400 acquires the voice data of the answer transmitted from the terminal 100 using the communication unit 202.
[0076] Next, in step S404, the data conversion unit 403 performs processing to convert the voice data of the response into text data, and acquires the target text.
[0077] In this example, speech data of the answer is acquired, but this is just an example, and target text may be acquired. In this case, instead of the processes of steps S403 and S404, a process of acquiring target text is executed.
[0078] Next, in step S405, the data conversion unit 403 converts the target text into a target vector by inputting the target text into the learning model stored in the storage unit 300. Details of this process will be described later with reference to FIG.
[0079] Next, in step S406, the uniqueness evaluation unit 401 uses the communication unit 402 to acquire a reference point corresponding to the subject indicated by the subject data acquired from the storage unit 300 in step S402.
[0080] Next, in step S407, the uniqueness evaluation unit 401 calculates the cosine similarity between the reference point and the target vector, and calculates a final evaluation value for evaluating the uniqueness of the answer from the calculated cosine similarity. Details of this process will be described later with reference to FIG.
[0081] Next, in step S408, the uniqueness evaluation unit 401 stores the calculated final evaluation value in the evaluation result table T2 in association with the target text, the user's identification information, the evaluation words, and the target vector. The final evaluation value is stored in the evaluation value field of the evaluation result table T2.
[0082] Next, in step S409, the uniqueness evaluation unit 401 generates display data for displaying a display screen showing the evaluation results of the answer, and transmits the generated display data to the terminal 100 using the communication unit 402. The terminal 100 receives this display data and displays the display screen on the display unit 105.
[0083] Fig. 8 is a flowchart showing the details of the processing of step S205 shown in Fig. 6 and step S405 shown in Fig. 7. In the explanation of Fig. 8, the reference text and the target text are collectively referred to as "text." Furthermore, the data conversion unit 203 and the data conversion unit 403 are collectively referred to as "data conversion unit A."
[0084] In step S2051, the data conversion unit A acquires text. Next, in step S2052, the data conversion unit A separates the read text into words. For example, morphological analysis is used for the separation.
[0085] Next, in step S2053, data conversion unit A refers to the excluded parts of speech list and deletes from the text any words that are not to be evaluated. This extracts evaluation words from the text. The evaluation words are nouns, adjectival verbs, verbs, and adjectives. The excluded parts of speech list stores excluded parts of speech other than these parts of speech. The excluded parts of speech list is stored in, for example, storage unit 300. Words with excluded parts of speech are deleted from the text data because such words do not contribute to the evaluation of uniqueness and may become noise.
[0086] Next, in step S2054, the data conversion unit A converts each evaluation word extracted from the text into a word vector. The reference vector is composed of a data set of word vectors calculated for each evaluation word. In this case, the reference calculation unit 201 may calculate the center of gravity of the multiple word vectors calculated for each evaluation word for each of the multiple reference vectors as a reference point. The target vector is composed of a data set of the multiple word vectors calculated for each evaluation word.
[0087] 9 is an explanatory diagram of the process of FIG. 8. In this example, "vacuum cleaner" is set as the subject. The user responds to the subject "vacuum cleaner" by inputting voice data of the answer using a microphone. In this example, the answer input is "play by straddling the handle."
[0088] The data conversion unit 403 converts the speech data of the response into text data. As a result, the target text D1 is obtained. Next, the data conversion unit 403 separates the target text D1 into words. As a result, the target text D1 is separated into words such as "pattern / of / part / straddle / play."
[0089] Next, the data conversion unit 403 deletes unnecessary words, which are words with excluded parts of speech, from the segmented target text D1. As a result, "pattern," "part," "straddle," and "play" are extracted as evaluation words from the target text D1.
[0090] Next, the data conversion unit 403 converts each evaluation word into a word vector using a vectorization method such as Word2vec, thereby obtaining a word vector for each evaluation word.
[0091] Fig. 10 is a flowchart showing the details of the processing of step S407 in Fig. 7. In step S4071, the uniqueness evaluation unit 401 acquires a reference score from the storage unit 300. Next, in step S4072, the uniqueness evaluation unit 401 calculates the cosine similarity between the reference score and the word vector for each evaluation word.
[0092] Next, in step S4073, the uniqueness evaluation unit 401 calculates an evaluation value for each evaluation word from the cosine similarity calculated for each evaluation word.
[0093] Next, in step S4074, the originality evaluation unit 401 calculates the maximum value of the evaluation values calculated for each evaluation word as the final evaluation value of the answer indicated by the target text.
[0094] 11 is an explanatory diagram of the process of FIG. 10. Unnecessary words are deleted from the segmented target text D1, and evaluation words are extracted from the target text D1. In this example, "pattern," "part," "straddle," and "play" are extracted as evaluation words.
[0095] The uniqueness evaluation unit 401 calculates a word vector for each of these four evaluation words. Map 1101 shows a vector space onto which the word vectors are mapped. Of these four evaluation words, the evaluation word "straddle" has the greatest distance from the reference point and therefore the greatest evaluation value. Therefore, the uniqueness evaluation unit 401 calculates the evaluation value of the evaluation word "straddle" as the final evaluation value. The word vector with the greatest evaluation value is an example of a vector that has the lowest similarity to the second vector among the multiple vectors corresponding to each of the multiple words included in the first answer text.
[0096] In this way, by identifying the evaluation word among multiple evaluation words that is the farthest from the reference point and evaluating the answer using the evaluation value of that evaluation word, it becomes easier to evaluate long answers.
[0097] Next, lighting control based on evaluation values will be described. Fig. 12 is a flowchart showing an example of lighting control processing. First, in step S501, the lighting control unit 501 of the lighting control server 500 obtains the final evaluation values for the answers from the storage unit 300 using the communication unit 502. Here, the final evaluation values for each of a plurality of answers proposed by users over a certain period of time on a certain topic are obtained.
[0098] Next, in step S502, the lighting control unit 501 calculates the number of responses whose final evaluation value is equal to or less than a threshold. The final evaluation value is an example of a plurality of scores derived for a plurality of responses acquired during a certain period.
[0099] Next, in step S503, the lighting control unit 501 determines whether the number of responses is equal to or greater than a predetermined number. If the number of responses is equal to or greater than the predetermined number (YES in step S503), the lighting control unit 501 generates a control command to change the illumination light and transmits the generated control command to the lighting device 600 via the communication unit 502. It is known that a user's creativity can be enhanced by lowering the color temperature of the illumination light and increasing the illuminance (see, for example, Non-Patent Document: Koichiro Tomimoto et al., "Office Lighting System Aimed at Improving Worker Creativity - Study on Lighting Conditions Suitable for Creative Office Scenes," Proceedings of the Kansai Branch Conference of the Japan Ergonomics Society, Volume: 2009, Pages: 171-174, Publication Date: December 5, 2009). Therefore, the lighting control unit 501 may generate a control command to lower the color temperature of the current illumination light by a predetermined level and increase the illuminance by a predetermined level. The condition that the number of responses is equal to or greater than a predetermined number corresponds to an example of a first condition indicating a low score.
[0100] On the other hand, if the number of answers is less than the predetermined number (NO in step S503), the process ends. In this case, after the next fixed period has elapsed, the process of FIG. 12 may be executed for answers created during the next fixed period. In FIG. 12, the lighting device 600 is controlled at fixed intervals, but this is only an example. Each time a user answers, a determination may be made as to whether or not to change the control of the lighting device. For example, if the final evaluation value for a certain answer is equal to or less than a threshold, the lighting control unit 501 may generate a control command and transmit the control command to the lighting device 600. In this case, the condition that the final evaluation value is equal to or less than a threshold is another example of the first condition.
[0101] If the originality of the user's answer is low or has decreased as a result of the process of FIG. 12, the lighting environment of the user can be controlled to encourage the user to create an answer with high originality.
[0102] Fig. 13 is a diagram showing an example of a display screen G1 displayed on the terminal 100 of the user U1 in step S409 of Fig. 7. The display screen G1 includes a user display field R1 that displays the user U1 who has provided an answer and the other users who have provided answers, and an evaluation result display field R2 that displays the evaluation results for the answers of the user U1 and the other users.
[0103] In this example, four users, including user U1, have provided answers, and therefore the images of these four users are displayed in the user display area R1. The evaluation result display area R2 displays each user's answer and the final evaluation value for the answer. If each user has provided multiple answers, the evaluation result display area R2 displays the multiple answers and the average of the final evaluation values for each answer. Alternatively, the evaluation result display area R2 may display the final evaluation value for each answer.
[0104] This allows user U1 to check what answers other users have given and how his / her answers and those of other users have been evaluated.
[0105] FIG. 14 is a diagram showing an example of a booth 1500. The booth 1500 is configured as a box that separates the space where the user is located from the external space. A lighting device 600 is installed on the ceiling of the booth 1500. Inside the booth 1500, a chair 1501 where the user sits and a desk 1502 where the user works are installed. A terminal 100 and a microphone 1503 are installed on the desk 1502. The user who answers uses the microphone 1503 or a keyboard to answer. The answers are evaluated by the originality evaluation server 400, and the lighting device 600 is controlled according to the evaluation results. For example, if there are many answers with low evaluation results and the evaluation results are unfavorable, the color temperature of the lighting device 600 is lowered and the illuminance is increased. This encourages the user to create highly original answers.
[0106] As described above, according to this embodiment, the uniqueness of an answer is evaluated not by the vector similarity between the topic and the target text indicating the answer to the topic, but by the vector similarity between the reference point indicating the standard for answers with low uniqueness associated with the topic and the target text. This reduces the risk that an answer with low uniqueness is erroneously evaluated as being highly unique. As a result, the uniqueness of an answer created from a certain topic can be evaluated more accurately.
[0107] The present disclosure can employ the following modifications.
[0108] (Variation 1) The uniqueness assessment unit 401 may evaluate answers using word vectors calculated for multiple corpora. For example, assume there are two corpora C1 and C2. In this case, the learning models include a learning model M1 corresponding to the corpus C1 and a learning model M2 corresponding to the corpus C2.
[0109] The reference points include two reference points P1 and P2 corresponding to corpora C1 and C2, respectively. The data conversion unit 203 inputs evaluation words included in the target text into learning models M1 and M2, respectively, and calculates word vectors V1 and V2 for each evaluation word for each of the corpora C1 and C2. The uniqueness evaluation unit 401 calculates an evaluation value between the word vector V1 and the reference point P1 for each evaluation word, and determines the maximum value K1 of the obtained evaluation values. Similarly, the uniqueness evaluation unit 401 calculates an evaluation value between the word vector V2 and the reference point P2 for each evaluation word, and determines the maximum value K2 of the obtained evaluation values. The uniqueness evaluation unit 401 then calculates the average of the maximum values K1 and K2 as the final evaluation value.
[0110] By calculating the final evaluation value using multiple corpora in this way, the fluctuations due to the corpora are suppressed, and the final evaluation value can be calculated with high accuracy. Here, for the convenience of explanation, the case of two corpora is exemplified, but the number of corpora may be three or more.
[0111] (Modification 2) A single word vector representing the characteristics of the target vector is calculated.
[0112] The following description will be given taking an example where two corpora C1 and C2 are used. Modification 2 is the same as Modification 1 up to the process of calculating word vectors V1 and V2 for each evaluation word using corpora C1 and C2. Here, it is assumed that there are three evaluation words. Therefore, the three word vectors V1 corresponding to corpus C1 are represented as V11, V12, and V13, and the three word vectors V2 corresponding to corpus C2 are represented as V21, V22, and V23.
[0113] The uniqueness evaluation unit 401 calculates one word vector V1' by calculating the element product of the word vectors V11 to V13, and calculates one word vector V2' by calculating the element product of the word vectors V21 to V23.
[0114] Next, the uniqueness evaluation unit 401 calculates the cosine similarity between the word vector V1' and the reference point P1, and calculates an evaluation value E1 of the answer corresponding to the corpus C1 from the calculated cosine similarity, and calculates the cosine similarity between the word vector V2' and the reference point P2, and calculates an evaluation value E2 of the answer corresponding to the corpus C2 from the calculated cosine similarity.
[0115] Next, the originality evaluation unit 401 calculates a final evaluation value by inputting the evaluation values E1 and E2 into a pre-created covariance structure analysis model. This covariance structure analysis model is a model that is trained to increase the weighting of corpora that calculate evaluation values close to human evaluations and decrease the weighting of corpora that calculate evaluation values that are far from human evaluations. By using the covariance structure analysis model, the weighting of evaluation values corresponding to corpora that produce evaluations that are far from human evaluations is decreased, while the weighting of evaluation values corresponding to corpora that produce evaluations close to human evaluations is increased, thereby calculating the final evaluation value. This allows for highly accurate evaluation of answers while suppressing evaluation variations due to corpora.
[0116] 1, the reference calculation server 200, the storage unit 300, the originality evaluation server 400, and the lighting control server 500 are configured as separate computers, but they may also be configured as a single computer. Each block of the reference calculation server 200, the storage unit 300, the originality evaluation server 400, and the lighting control server 500 may be provided in the terminal 100.
[0117] (Variation 4) In the above embodiment, the uniqueness evaluation unit 401 acquires the target vector converted by the data conversion unit 403, but this is just one example, and the target vector calculated by a device provided outside the information processing system 1 may be acquired from that device. Similarly, the reference calculation unit 201 acquires the target vector converted by the data conversion unit 203, but this is just one example, and the reference calculation unit 201 may acquire the target vector calculated by a device provided outside the information processing system 1. That is, acquiring the first vector and the second vector in the present disclosure includes not only calculating the first vector and the second vector, but also acquiring the first vector and the second vector from an external device.
[0118] (Modification 5) In the above embodiment, cosine similarity is used as the similarity, but the present disclosure is not limited to this. For example, Pearson's correlation coefficient may be used as the similarity.
[0119] The present disclosure is useful in the technical field of improving user work efficiency.
[0120] 1: Information processing system 100: Terminal 101: Input unit 102: Communication unit 103: Screen generation unit 104: Screen control unit 105: Display unit 200: Reference calculation server 201: Reference calculation unit 202: Communication unit 203: Data conversion unit 204: Subject setting unit 300: Storage unit 301: Storage unit 302: Communication unit 400: Originality evaluation server 401: Originality evaluation unit 402: Communication unit 403: Data conversion unit 404: Subject setting unit 500: Lighting control server 501: Lighting control unit 502: Communication unit 600: Lighting device
Claims
1. An information processing method on a computer, comprising: obtaining a first vector corresponding to a first answer text indicating an answer to a topic; obtaining a second vector corresponding to a second answer text indicating an answer associated with the topic; and evaluating the uniqueness of the answer indicated by the first answer text based on the degree of similarity between the first vector and the second vector.
2. The information processing method according to claim 1, wherein the second answer text includes a plurality of texts, and the second vector is a vector corresponding to a center of gravity position determined from a plurality of vectors generated by converting each of the plurality of texts into a vector.
3. The information processing method according to claim 1, wherein the similarity is a cosine similarity derived based on the first vector and the second vector.
4. The information processing method according to claim 1, wherein the first answer text is obtained by converting voice data based on a user's speech into text data.
5. The information processing method according to claim 1, wherein evaluating the uniqueness of the answer includes deriving a score according to the uniqueness of the answer.
6. The information processing method according to claim 5, further comprising the step of controlling a lighting device that illuminates the vicinity of the user to lower the color temperature of the illumination light and to increase the illuminance when a first condition indicating that the score is low is satisfied.
7. The information processing method according to claim 6, further comprising determining that the first condition is satisfied when the score is smaller than a threshold value.
8. The information processing method of claim 6, wherein obtaining the first vector includes obtaining a plurality of first answer texts; evaluating the uniqueness of the answers includes deriving a plurality of scores for a plurality of answers corresponding to each of the plurality of first answer texts obtained during a certain period of time; and determining whether the first condition is met if a predetermined number or more of answers have a score below a threshold during the certain period of time among the plurality of scores.
9. The information processing method according to claim 1, wherein the first vector is a vector that has the lowest similarity to the second vector among a plurality of vectors corresponding to each of a plurality of words included in the first answer text.
10. The information processing method according to claim 1, further comprising the step of: displaying on a display a screen for displaying the evaluation result of the originality of the answer.
11. The information processing method of claim 1, wherein the first vector includes a plurality of word vectors calculated for each of a plurality of words contained in the first answer text for each of a plurality of corpora, the second vector includes a plurality of reference points corresponding to the plurality of corpora, and the evaluation of uniqueness includes calculating a plurality of evaluation values regarding uniqueness corresponding to the plurality of corpora by comparing the plurality of word vectors with the plurality of reference points for each of the plurality of corpora, and calculating a final evaluation value from the plurality of evaluation values.
12. An information processing system having a processor, wherein the processor performs the following processes: acquire a first vector corresponding to a first answer text indicating an answer to a topic; acquire a second vector corresponding to a second answer text indicating an answer associated with the topic; and evaluate the uniqueness of the answer indicated by the first answer text based on the degree of similarity between the first vector and the second vector.
Citation Information
Patent Citations
Questionnaire conducting program, questionnaire conducting device, and questionnaire conducting method
JP2004021433A
Information processing system, information processing method and program
JP2020170460A
