JP2015179212A - Content presentation system, method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a content presentation system capable of automatically acquiring information related to a state suitable for presenting content such as e-learning to a user.SOLUTION: An environment information acquisition part 22 acquires the environment information of a user terminal. A presentation part 33 presents content to a user, and a reception part 34 receives a response from the user. A management part 32 manages the content, and obtains an evaluation result for the receive response. A creation part 4 creates training data by associating the acquired environment information with the obtained evaluation result. A construction part 5 constructs a prediction model of whether or not the environment information corresponds an appropriate state for presenting the content to the user on the basis of the training data.

Description

  The present invention automatically detects a state suitable for a user to tackle learning and the like when presenting content for learning in e-learning (electronic learning) using gap time, and then The present invention relates to a content presentation system, method and program capable of presenting content at a certain time.

  As a technique for automatically presenting content during a user's gap time, for example, there is a technique disclosed in Patent Document 1. Patent Document 1 discloses that a boarding time is set as a gap time, and scheduling for presenting a plurality of contents according to the length of the gap time is performed. In this technique, the current location and the destination are input to the car navigation system to determine the boarding time, and scheduling is performed for the known gap time as the boarding time.

JP 2010-39805 A (Content Scheduling Device)

  In recent years, with the widespread use of smartphones, situations where information processing terminals are always carried are becoming more common. In such situations, it is possible to effectively present learning content in e-learning using gap time. desired.

  However, when considering the effective presentation of learning content in e-learning or the like using gap time as described above, the technique of Patent Document 1 has the following problems.

  That is, the technique of Patent Document 1 has a problem that the clearance time for learning is limited to the boarding time. Also, for example, if the train is crowded even if it is a gap time while getting on the train, it is not necessarily a time suitable for learning, and even if the learning content is presented at such time, the user It is not possible to learn effectively. However, the technique of Patent Document 1 has a problem that it is impossible to determine a time inappropriate for such learning.

  The present invention has been made in view of the above-described problems of the prior art, and an object thereof is to provide a content presentation system, method, and program capable of automatically obtaining information related to a state suitable for presenting content to a user. To do.

  To achieve the above object, the present invention provides a content presentation system, an environment information acquisition unit that acquires environment information of a user terminal, a presentation unit that presents content to the user, and a response to the presented content from the user And managing the content to be presented and receiving the response, and creating the training data by associating the acquired environment information and the obtained evaluation result with the managing unit for obtaining the evaluation result for the received response And a constructing unit that constructs a prediction model based on the training data to determine whether the environmental information corresponds to an appropriate state for presenting content to the user. First feature.

  Further, the present invention provides the content presentation system according to the first feature, wherein the presentation unit applies the prediction model constructed by the construction unit to the environment information obtained by the environment information acquisition unit, and A second feature is to present content to the user when it is determined that the environmental information corresponds to an appropriate state.

  In addition, the present invention is a content presentation method that periodically obtains environment information of a user terminal, periodically presents content to a user for a previously managed content, and from the user. A step of receiving a response to the presented content, a step of creating training data by linking the acquired environmental information and the obtained evaluation result, and the environmental information presenting the content to the user based on the training data And a step of constructing a prediction model as to whether or not it corresponds to an appropriate state for performing the third feature.

  Furthermore, the present invention has a fourth feature that it is a content presentation program that causes a computer to function as the content presentation system.

  According to the first, third, or fourth feature, a prediction model is automatically constructed as to whether the environmental information in the user terminal corresponds to an appropriate state for presenting content to the user. Therefore, it is possible to automatically obtain information on a state suitable for presenting content to the user.

  According to the second feature, when the user terminal is in a state suitable for presenting content to the user, the content can actually be automatically presented.

It is a functional block diagram of the content presentation system which concerns on one Embodiment. It is a figure which shows the example of the memorization problem content of an English word as an example of a content. It is a flowchart showing each phase of operation | movement of the content presentation system which concerns on one Embodiment. It is a figure which shows the example of the training data which a preparation part produces. It is an example of the flowchart of training data creation in 1st embodiment. It is an example of the flowchart of training data creation in 2nd embodiment. FIG. 5 is a diagram illustrating an example in which conversion such as normalization is performed in order to use the training data in FIG. 4 as learning data for machine learning. It is an example of the flowchart in an operation phase. It is an example of the flowchart which performs training data creation and prediction model update in an operation phase by creating training data in a first embodiment. It is an example of the flowchart which performs training data creation and prediction model update in an operation phase by creating training data in a second embodiment.

  FIG. 1 is a functional block diagram of a content presentation system according to an embodiment. The content presentation system 10 includes a sensor 21 and an environment information acquisition unit 22, and a content database (content DB) 31, a management unit 32, a presentation unit 33, A reception unit 34, a creation unit 4, and a construction unit 5 are provided.

  The content presentation system 10 may be realized entirely by a user terminal such as a smartphone or a head-mounted display type terminal carried by the user, or a part of functional units may be provided by a server on a network separate from the user terminal. It may be realized. Even when a part of the functional units is assigned to the server or the like, it is preferable that at least the sensor 21, the presenting unit 33, and the receiving unit 34 are provided on the user terminal side.

  The function of each part in FIG. 1 is as follows.

  The sensor 21 includes various devices such as sensors provided in the user terminal, obtains the state of the user terminal as environment information, and passes it to the environment information acquisition unit 22. The environment information acquisition unit 22 passes the acquired environment information to the creation unit 4 in the “data collection phase” described later, and presents the presentation unit 33 (or the presentation unit 33 and the creation unit 4 in the “operation phase” described later). ). The device such as the sensor and the acquired environmental information can include any combination of [1] to [9] below.

  That is, [1] location information acquired from GPS (Global Positioning System) sensor, [2] date and time information acquired by clock, [3] environmental sound information acquired by microphone, [4] acquired by camera [5] Illuminance information acquired by the illuminance sensor, [6] User terminal orientation information acquired by the gravity sensor, [7] User terminal acceleration information acquired by the acceleration sensor, [8] Temperature Temperature information acquired by a meter, [9] humidity information acquired by a hygrometer, and the like can be acquired as environmental information.

  Note that the date / time information in [2], temperature information in [8], humidity information in [9], etc. are acquired from the server holding the information via the network, not from the sensor 21 provided in the user terminal. May be. At this time, with respect to [8] and [9], for example, the position information can be acquired by transmitting it to a server on the network (a server holding the distribution information of temperature information and humidity information). Similarly, the location information may be transmitted to a server on the network, and the weather information at the location where the user terminal exists may be included in the environmental information to be acquired. The image information of [4] may be acquired as environmental information after extracting a known feature amount (SIFT feature amount or the like). The environmental sound information of [3] may also be acquired as environmental information after being converted into a feature quantity such as a component frequency.

  The respective units 31 to 34 responsible for providing contents to the user and managing the history and the like are as follows.

  The content DB 31 stores various types of content provided to the user. In general, the content can be evaluated by giving a problem to the user and obtaining an answer or other response from the user. The content is created in advance by hand or automatically or by other methods.

  The content may be, for example, learning content or quiz content such as an English word memorization problem or a history year memorization problem, and in this case, the user's degree of understanding is evaluated by the content. In addition, the content may be game content or puzzle content that requires a specific solution, such as Go, Shogi, chess, and the like, and in this case also, a method for dealing with a specific situation in the game or puzzle Regarding others, the user's understanding level or proficiency level is evaluated.

  As described above, the user's “response” to the content includes an answer, a solution, a procedure, an operation content in a game, and other various types according to the specific content. In the following explanation, from the viewpoint that it is easy to understand as a specific example for explanation, explanation is given using “answer” as an example of “response”, but the explanation is not limited to “answer” but is general. The same holds true for “response”.

  FIG. 2 is a diagram illustrating an example of an English word memorization problem content as an example of the content. A column C1 is a screen example of content when presenting a problem to the user, and a column C2 is a screen example of content when presenting the correct answer to the user to the user. In the example of the column C2, the correct answer is shown as it is, but the correct answer may be presented in a form such as correcting the answer input by the user.

  The presentation unit 33 issues a request to the management unit 32 at every predetermined timing to receive the content (content selected by the management unit 32 as described later) and present it to the user. The receiving unit 34 receives an answer from the user for the presented content. In addition, after receiving the answer from the user at the reception unit 34, the presentation unit 33 presents the correct answer in the content (or an exemplary procedure according to the content) to the user. When the content is a game or the like and there is no specific correct answer, the presentation of the correct answer may be omitted.

  Here, the presentation unit 33 can be realized by a display provided in the user terminal as long as the content is displayed, and can be realized by a speaker provided in the user terminal if the content is audio. A combination of these can also be realized. In the case of displayed content, a see-through type head-mounted display may be superimposed on a scene viewed by the user using a known AR (augmented reality) technology. Similarly, when the content is other multimedia content, the presentation unit 33 can be realized by a known compatible device. Similarly, the reception unit 34 can also be realized by a keypad, a touch panel, a microphone, an acceleration sensor, or other known corresponding devices provided in the user terminal according to the content such as an answer to the content.

  The management unit 32 realizes content presentation to the user in the presentation unit 33 by selecting and presenting the content to be presented and receiving the answer from the content DB 31 in response to a request from the presentation unit 33 At the same time, the reception unit 34 evaluates the answer received from the user for the content, and assigns a correct answer / incorrect answer or a score.

  The evaluation can be automatically performed by the management unit 32 by comparing a predetermined correct answer in the content with an answer obtained from the user. For example, if the content allows the user to select the correct answer from the options, it may be automatically determined whether the option selected by the user matches the correct option. Further, if the content is a game, a puzzle, or the like, a score or the like may be automatically assigned according to the score assignment method, rules, or the like. Further, depending on the type of content, the evaluation may be separately given manually. For example, if the content requires an answer in a descriptive format, the management unit 32 may transmit the answer obtained from the user to the grader on the network, and return the evaluation result from the grader.

  Thus, when the reception unit 34 receives an answer from the user for the content presented to the user by the presentation unit 33, the management unit 32 evaluates the answer, and the evaluation result is sent to the creation unit 4 via the reception unit 34. Is passed. In the series of processes, the management unit 32 manages the content presented to the user and the evaluation result as a history including the presented date and time.

  Note that when the content to be presented to the user from the content DB 31 is selected by the management unit 32, the content is selected from the types of content determined in advance by the user. For example, if the user has installed an English word learning app on the user terminal to carry out e-learning, or if it is registered on an English word learning site, content corresponding to the app or site (multiple English words What is necessary is to select from the content of memorization problem content).

  Further, regarding the interface for presenting the content to the user by the presenting unit 33 and the interface for receiving the answer from the user by the accepting unit 34, in addition to those illustrated in FIG. Various known interfaces can be used.

  The creation unit 4 creates training data by combining the environment information obtained from the environment information acquisition unit 22 and the evaluation result of the user's content answer obtained from the reception unit 34, and passes the training data to the construction unit 5 . Based on the training data, the construction unit 5 obtains a prediction model for predicting whether or not the environment information of the user terminal obtained by the environment information acquisition unit 22 corresponds to an appropriate state for presenting content to the user. Then, the obtained prediction model information is passed to the presentation unit 33.

  Thus, in the “data collection phase” to be described later, the presentation unit 33 is a stage before the prediction model is created, and thus the content is periodically presented without referring to the prediction model. On the other hand, in the “operation phase” to be described later, since the prediction model has already been created, the presentation unit 33 refers to the prediction model, and the environment information obtained from the environment information acquisition unit 22 sends the content to the user. When it is determined that the state is suitable for presentation, the content is presented.

  The function of each part in FIG. 1 has been described above. FIG. 3 is a flowchart showing each phase of the operation of the content presentation system 10 according to an embodiment. With reference to FIG. 3, the operation of each part in FIG. 1 (time-series relationship of the function of each part in FIG. 1) will be described.

  The first phase P1 is a “data collection phase”. In the data collection phase, the presentation unit 33 presents content to the user at each predetermined timing, the reception unit 34 receives an answer to the content from the user, the management unit 32 gives an evaluation to the answer, and the presentation When the environment information acquisition unit 22 acquires environment information in the user terminal when the answer is received, the creation unit 4 creates training data.

  In the first embodiment, the training data is information obtained by associating the evaluation result [t1] of the answer from the user and the environment information [t1] at the date and time t1 when each content is presented {evaluation result [t1], environment The information [t1]} is created by aggregating it over the contents that have been presented and obtained answers. By continuing the counting, the creation of training data is completed when training data for a predetermined number of samples N is obtained.

  FIG. 4 is a diagram illustrating an example of training data created by the creation unit 4 in the data collection phase. Here, an example in which environmental information is configured by time, day of the week, temperature, and sound is shown, and whether the answer from the user when presenting the environmental information and each content is correct or incorrect (ie, The training data as shown in the figure is created as information in which the evaluation results of the answers are associated with each other.

  In addition, as a second embodiment in which training data is created by associating environmental information and an evaluation result, the following may be performed. In other words, if the answer obtained by presenting the content to the user for the first time at a certain date and time t2 is incorrect, environmental information [t2] is acquired, and the time and date after that time t3 (t3> t2) Again, the content that was incorrect in the initial presentation is presented, and the evaluation result [t3] of the answer obtained from the user is obtained. In this way, the information {evaluation result [t3], environment information [t2]} obtained by associating the evaluation result [t3] with the environmental information [t2] is totaled over each content, and the predetermined number of samples N When the training data is obtained, the creation of the training data is completed.

  Note that the format of the training data itself is common to the second embodiment and the first embodiment, and the training data as illustrated in FIG. 3 is obtained also in the second embodiment.

  The second embodiment is based on the following assumption. That is, for the content that was presented for the first time at the date and time t2, the user would learn by looking at the correct answer, but if the learning was made appropriately at the date and time t2, the same content The correct answer should be obtained at the date and time t3 presented next, while if the learning was inappropriate at the date and time t2, the correct answer should still be incorrect at the next date and time t3 presented. It is an assumption.

  Thus, in the second embodiment, by detecting whether or not the environmental information [t2] at the date and time t2 that was incorrect in the first time was suitable for learning, as an evaluation result [t3] at a later date t3, Create training data. The second embodiment is suitable when the content is for learning unknown knowledge, such as memorizing English words shown in FIG.

  On the other hand, the first embodiment is suitable for content in which the evaluation result of the answer to the content does not depend on the presence or absence of the user's knowledge but depends only on the degree of user concentration at the time t1. For example, the first embodiment is suitable for content such as a puzzle or a game in which a score is determined according to the user's procedure and skill, not necessarily one correct answer.

  FIG. 5 is an example of a flowchart for creating training data in the first embodiment. In the following description, particularly in the description of the flow shown in FIGS. 5, 6, and 8 to 10, each content has some problem for the user from the viewpoint that it is easy to understand as a specific example for explanation. However, the content is not limited to what is presented, but may be other content (content that requests a response from the user). Exactly the same processing is possible. In addition, the case of general content that is not limited to “question” will be mentioned as appropriate.

  In step S10, the management unit 32 refers to the content DB 31, creates a question list to be provided to the user, and then proceeds to step S11. Here, when the user takes e-learning or the like using the content presentation system 10, a question list is created in accordance with the content to be taken. At the time of the creation, the order of the questions may be determined in the question list. In the case of a general case that is not limited to “question”, a “content list” is created.

  In step S11, the presenting unit 33 determines whether or not the time at which the question should be given to the user has been reached. If it has reached, the process proceeds to step S12, and if not, the presenting unit 33 proceeds to step S11. Wait at

  It should be noted that the time for obtaining the judgment from step S11 to step S12 (the time of the question timing) may be set at a predetermined interval such as every 10 minutes. The predetermined interval may be determined to vary depending on the day of the week or the time zone. For example, the night when the user is considered to be sleeping may not be included in the question timing time and may continue to wait in step S11.

  In step S12, the presentation unit 33 provides questions to the user (generally, content is provided), and then proceeds to step S13. Here, as the question to be given, the management unit 32 selects any one question from the question list held at the time, and the presentation unit 33 presents the selected question to the user. Alternatively, if the order of the questions is defined in advance in the question list creation in step S10, the questions according to the order may be selected.

  In step S13, it is confirmed whether or not an answer has been received from the user in the reception unit 34 within a predetermined time from the problem presentation time for the problem presented to the user in step S12, and the answer has been obtained. If it has not been obtained, the process returns to step S11. When an answer is obtained, the presentation unit 33 may further present the correct answer of the problem as described above.

  In step S14, the environment information acquisition unit 22 acquires the environment information of the user terminal at the time when the answer is received, and then the process proceeds to step S15. The acquisition of the environment information is not limited to the time when the answer is received, but may be acquired at any time from the time when the problem is presented to the time when the answer is received.

  In step S15, the management unit 32 evaluates the answer confirmed to have been received from the user in the most recent step S13 and obtains an evaluation result. After the training data is generated by associating with the environment information acquired in the latest step S14, the process proceeds to step S16. In this way, every time the process reaches step S15, the creation unit 4 newly generates training data, whereby the training data is accumulated.

  In step S16, the management unit 32 deletes the question that has been answered by the question from the question list, so that it is excluded from the questions to be answered for the user at a subsequent time point. Proceed to step S17.

  In step S17, it is determined whether the training data created and accumulated in the creation unit 4 has reached a predetermined number of samples (that is, whether the number of times step S15 has been executed has reached a predetermined number), If it has reached, the flow of FIG. 5 is completed assuming that the training data creation is completed, and if not reached, the training data creation is not completed, and the process returns to step S11.

  FIG. 6 is an example of a flowchart for creating training data in the second embodiment. In FIG. 6, steps S20 to S23 and S27 are the same as steps S10 to S13 and S17 of the first embodiment described in FIG. S23 and S27 are as follows.

  In step S23, as in step S13, a determination is made as to whether an answer has been obtained from the user for the presented question. If an affirmative determination is made, the process proceeds to step S24, and a negative determination is made. If YES, return to step S21. In step S27, as in step S17, a determination is made as to whether training data creation has been completed by accumulating a predetermined number of training data. If the determination is affirmative, the flow of FIG. Returns to step S21.

  In step S24, the management unit 32 determines whether or not the problem for which the answer has been obtained is the first question for the user for the problem for which the reception of the answer from the user has been confirmed in the most recent step S23. Since the management unit 32 manages a history of a series of questions for the user together with the question list, the judgment in step S24 can be made by referring to the history.

  In the history, questions that have been answered are managed as questions, and questions that have not been answered (questions that have been returned from step S23 to step S21) are managed as not being questioned. Accordingly, even for questions that have already been answered in the past but have not been answered, the first question is determined in step S24 when the answer is obtained for the first time.

  If the determination in step S24 is affirmative and is the first question, the process proceeds to step S25. On the other hand, if the determination is negative and the question is a second or subsequent question, the process proceeds to step S30.

  In step S25, the management unit 32 determines whether or not the answer from the user to the problem determined to be the first question in the most recent step S24 is a correct answer. If the answer is correct, step S26 If the answer is incorrect, the process proceeds to step S28.

  In step S26, the management unit 32 deletes the question for which the answer is determined to be correct in the most recent step S25 from the managed question list, and excludes it from the questions to be asked at a later time, Proceed to step S27.

  In step S28, the environment information acquisition unit 22 acquires the environment information of the user terminal at the time point when the answer is determined to be incorrect in the most recent step S25, and the process proceeds to step S27. The environmental information may be acquired at any point in time from when the first question is given to when it is answered.

  On the other hand, in step S30, which is determined not to be the first question in step S24, the management unit 32 managing the question history as described above, the problem that is not the first question is given to the user. If it is the second question, the process proceeds to step S31. If not, that is, if it is the third or later question. Proceed to step S35.

  In step S31, the management unit 32 determines whether or not the user's answer to the question determined to be the second question in the most recent step S30 is a correct answer. If it is determined that the answer is incorrect, the process proceeds to step S34.

  In step S32, the creation unit 4 creates training data according to the problem determined as the second question in the most recent step S30, and then the process proceeds to step S33. Here, the training data is created by associating the environmental information acquired in step S28 with the second evaluation result (correct answer) at the time of the first question in which the answer is incorrect.

  In step S33, the management unit 32 removes from the question list the questions for which the second question is given and the answer from the user is correct, thereby removing the question from the questions to be asked at the subsequent time point and proceeds to step S27. .

  In step S34, the creation unit 4 creates training data according to the problem determined as the second question in the most recent step S30, and then the process proceeds to step S27. Here, training data is created by associating the environmental information acquired in step S28 with the second evaluation result (incorrect answer) at the time of the first question when the answer is incorrect. .

  On the other hand, in step S35, where it is determined that the question is not the second question in step S30 (that is, the third and subsequent questions), the answer obtained from the user for the third and subsequent questions is the correct answer. The management unit 32 determines whether or not there is a correct answer. If the answer is correct, the process proceeds to step S36. If the answer is incorrect, the process proceeds to step S27.

  In step S36, the manager 32 removes the question that is given as the third or later question and the answer from the user is correct from the question list by removing it from the question list. Proceed to

  The steps in the flow of FIG. 6 have been described above, and supplementary items (1) to (3) will be described.

  (1) As described above with reference to the second embodiment, the flow of FIG. 6 is suitable for content that allows the user to perform memorization learning and the like. That is, steps S26, S33, and S36 for deleting from the question list by the management unit 32 are branched when the answer to the question is correct in the most recent determination steps S25, S31, and S35, respectively. Arranged as steps. Therefore, each question remains in the question list until the user gives a correct answer, and there is no chance that the question will be given until the correct answer is obtained. For this reason, the flow of FIG. 6 is suitable for memorization learning and the like.

  In order to obtain the same effect as described above, in the flow of FIG. 5, the deletion from the question list in step S16 is performed only when the answer from the user to the question is correct. Also good.

  (2) When the answer (response) to the content cannot give a correct / incorrect binary classification, for example, the content is a game or a puzzle, and the answer (response) is evaluated as a score. 6 is applied by assuming that a high score is “correct” and a low score is “incorrect” by adding a predetermined threshold judgment to the score, etc. .

  (3) When performing the question in step S22, the management unit 32 selects a question to be asked from the question list. At this time, the question that has already been given and obtained an answer is an incorrect question. If it is in the list, it may be excluded from the selection target for the question until a certain period of time elapses after the latest incorrect answer is obtained. In this way, after a certain amount of time has passed, the question of the incorrect answer is given again, so that the degree of learning enhancement and the degree of memory fixation in the user can be more reliably examined. Similarly, when a certain period of time has elapsed since a certain problem was incorrect, the problem that was the incorrect answer may be given.

  Returning to the description of FIG. 3, when the creation of training data by the creation unit 4 is completed by the first and second embodiments as described above, the data collection phase as the first phase P1 is completed, and the second phase P2 Proceed to

  The second phase P2 is a “training phase” in which the construction unit 5 uses the training data created in the first phase P1 to provide content to the user among the environmental information acquired by the environmental information acquisition unit 22. Build a model that predicts what is appropriate for your presentation. For the prediction model construction, the following first and second embodiments are possible. (The first and second embodiments of training data creation described above and the first and second embodiments of the prediction model construction are independent of each other.)

  In the first embodiment of the prediction model construction, referring to the training data given in a form that links the environmental information and the evaluation result, the environmental information when the evaluation result can be determined to be good is sent to the user as content. A predictive model can be constructed that corresponds to a state suitable for provision.

  At this time, whether or not the evaluation result is good is determined when the evaluation result is given as a correct answer / incorrect answer, and when the evaluation result is given as a score, etc. If there is, it may be determined that the case where a high score is determined based on the threshold condition is good as described in the supplementary item (2) of the flow of FIG. When the environmental information in the evaluation result determined to be good includes continuous information such as a numerical value of temperature, the predetermined range including the numerical value determined to be good may correspond to a suitable state.

  Furthermore, in order to determine whether or not the state is suitable for content presentation by applying a prediction model for arbitrary environmental information (environment information as a variable variable), the following first method or The second method may be used. In this way, a prediction model is constructed that outputs whether it corresponds to a state suitable for content presentation using environmental information as an input.

  That is, in the first method, the environment information corresponding to the evaluation result determined to be good as described above in the training data (environment information linked to the evaluation result determined to be good on the training data) is completely stored. If they match, it may be determined that a suitable state is met. Alternatively, as a second method, a case where a certain limited attribute in the environment information matches based on a rule manually determined in advance may be determined to correspond to a suitable state. For example, in the example of FIG. 3, the environment information includes four attributes of time, day of the week, temperature, and sound. Of these, determination may be made by matching only two attributes of time and day of week.

  In the second embodiment for constructing the prediction model, a prediction model may be constructed by machine learning. That is, by using the training data given in a form that links the environmental information and the evaluation result as learning data, by performing machine learning such as a well-known SVM (support vector machine), the environmental information is input, A prediction model may be constructed by constructing a discriminant that outputs whether or not a state in which a good evaluation result is obtained.

  When performing machine learning, the environment information in the training data is normalized according to a predetermined rule and converted into a feature form input to a discriminant constructed by machine learning. In addition, after determining whether or not the evaluation results in each training data may be the same as in the first embodiment, a case where it is determined to be good corresponds to a positive example, and a case where this is not the case is a negative example. Classify training data as applicable.

  FIG. 7 shows an example of the training data in FIG. 4 which is normalized for machine learning and converted into a feature form, and a positive example in which the teacher signal is “1” and the teacher signal is “0”. It is a figure which shows the example classified into the negative example. In FIG. 7 and FIG. 3, each row (the second and subsequent rows excluding the column of the heading row) represents the content in the same training data corresponding to each other.

  In FIG. 7, the “correct answer” training data in FIG. 4 is classified as a positive example of the teacher signal “1”, and the “incorrect answer” training data in FIG. 4 is classified as a negative example of the teacher signal “0”. . Further, in FIG. 7, “time” in FIG. 4 is discretized to become each “time zone” as a nominal scale, and similarly, “day of the week” is also set to each “day of the week” as a nominal scale. In FIG. 7, “temperature” and “sound” in FIG. 4 are normalized. The normalization method includes a method of dividing by the maximum value of each feature and a method of standardization using the average and standard deviation of each feature.

  Returning to FIG. 3, when the second phase P2 of the training phase is completed as described above, the process proceeds to the third phase P3. The third phase P3 is the “operational phase”, and is suitable for the presentation unit 33 to present the content to the user using the prediction model constructed in the training phase. The content is actually presented to the user when it is determined that the condition is met.

  For this reason, the presentation unit 33 acquires environmental information from the environmental information acquisition unit 22 at regular intervals, and determines that the result of applying the prediction model to the acquired environmental information is in a suitable state. A content request is issued to the unit 32 to receive the content, and the content is presented to the user.

  In this way, in the operation phase, from the standpoint of the user who owns the user terminal, content is automatically presented when it is estimated that the user is in a state suitable for performing learning and the like. Etc. can be carried out effectively.

  FIG. 8 is an example of a flowchart in the operation phase. In step S40, as in step S10 of FIG. 5, the management unit 32 creates a question list, and then the process proceeds to step S41. If the content of the content is something that the user performs knowledge-fixed learning, such as memorization learning, the question list will be based on questions that have not been answered or have not been answered in the “training phase”. It is preferable to create.

  In step S41, a time counter t for managing a time interval for making a regular determination as to whether or not content presentation should be performed is set to an initial value 0, and then the process proceeds to step S42. In the counter t, the elapsed time from the time when the initial value 0 is set is updated and stored in real time.

  In step S42, it is determined whether the elapsed time t managed by the counter t has reached a constant value T. If it has reached, the process proceeds to step S43, and if not, the process proceeds to step S49. In step S49, the counter t is updated and incremented by a predetermined unit time, and the process returns to step S42. Thus, step S42 is repeated for each unit time, and the arrival of the elapsed time t to the fixed time T is awaited.

  In step S43, the environment information acquisition unit 22 acquires the environment information in the user terminal from the sensor 21 and passes it to the presentation unit 33, and then the process proceeds to step S44. In step S44, the presentation unit 33 applies the prediction model constructed by the construction unit 5 to the environment information at the current time obtained from the environment information acquisition unit 22, so that the environment information provides questions to the user. Then, it is determined whether it is suitable for learning (providing content) or the like, and the process proceeds to step S45.

  In step S45, cases are classified according to the determination result in the latest step S44. If it is determined that the environment information is suitable for the user to perform the questions, the process proceeds to step S46, and the environment information is not suitable. If determined, the process returns to step S41.

  In step S46, after the management unit 32 selects a question to be asked from the question list, the presentation unit 33 gives a question to the user, and the process proceeds to step S47. The selection in step S46 may be random as described for step S12 in FIG. 5, or may be performed in the order in which the order of questions is set in the question list.

  In step S47, the management unit 32 deletes the question that was asked in the latest step S46 from the question list, and is excluded from the questions that are given to the user for the user at a later time. Then, the process proceeds to step S48.

  In step S48, it is determined whether or not the question list currently managed by the management unit 32 is empty, that is, whether or not all questions have been completed. This flow ends, and if it is not completed, the process returns to step S41.

  As described above, according to the present invention, it is possible to automatically construct a prediction model that predicts whether environmental information in a user terminal corresponds to a state suitable for presenting content for the user to perform learning or the like. At this time, by confirming the response from the user while presenting the content to the user, by accumulating the training data by acquiring the environment information of the user terminal when there is a response, the prediction model is used by using the training data Can be built automatically. Moreover, by detecting a state suitable for the user to perform learning and the like using the constructed prediction model, and presenting the content when the user is in the suitable state, learning and the like in the gap time can be efficiently performed. It becomes possible to do.

  Hereinafter, supplementary items (Supplement 1) to (Supplement 4) in the present invention will be described. In addition, the said supplementary matter is a supplementary matter from a viewpoint of clarifying the flow of description, and is contained in embodiment of this invention.

  (Supplement 1) In the third phase P3, which is the “operational phase” shown in the flow of FIG. 3, by continuing the accumulation of training data, the prediction model for predicting a state suitable for content presentation may be updated. Good. That is, in the operation phase, the data collection phase and the training phase may be continued to update the prediction model to be more suitable for the situation of the user.

  FIG. 9 is an example of a flowchart for creating training data and updating a prediction model in the operation phase by the training data creation method according to the first embodiment described in FIG.

  In FIG. 9, steps S50 to S57 are the same as steps S10 to S17 of FIG. 5 to which corresponding numbers (step numbers differ by 40) are assigned respectively, and thus redundant description is omitted. However, steps S51 and S57 in FIG. 9 are steps modified from the corresponding steps S11 and S17 in FIG. 5 as follows.

  In step S51 of FIG. 9, it is determined whether or not the timing of the question has been reached, but this determination does not wait for the elapse of a fixed time as in step S11 of FIG. The environmental information is obtained from the acquisition unit 22, and it is determined whether or not the result of applying the prediction model held at the time to the environmental information is in a state suitable for content presentation. That is, the processing corresponding to the loop portion of steps S41 to S45 and S49 in FIG. 8 is performed in step S51 in FIG.

  In step S57 of FIG. 9, unlike determining whether the training data creation is completed in step S17 of FIG. 5, it is determined whether or not the question list is empty, that is, in the question list created in step S40. Whether or not all questions have been completed is determined. If the list is empty and the questions have been completed, the process proceeds to step S58; otherwise, the process returns to step S51.

  In step S58 of FIG. 9, the construction unit 5 newly constructs a prediction model using the training data newly obtained by the flow of FIG. 9, and a fluctuation of a certain amount or more from the already constructed prediction model. If so, the newly constructed prediction model is adopted for future use. Or you may make it always employ | adopt a new prediction model irrespective of some fluctuations. When the operation phase is further continued with the new prediction model, the flow in FIG. 9 may be continued under the new prediction model.

  In step S58, the construction unit 5 may construct a new prediction model using only the training data newly obtained by the flow of FIG. 9, or the training data already obtained in the training phase may be used. The construction unit 5 may construct a new prediction model in a form that is additionally used. In addition, a new prediction model may be constructed using only training data obtained in the past for a predetermined period (training data obtained in the latest past) using the time point at which step S58 is reached as a reference.

  FIG. 10 is an example of a flowchart for creating training data and updating a prediction model in the operation phase by the training data creation method according to the first embodiment described in FIG.

  Since steps S60 to S68 and S70 to S76 in FIG. 10 are the same as steps S20 to S28 and S30 to S36 in FIG. Is omitted. However, steps S61 and S67 in FIG. 10 are as follows, unlike the corresponding steps S21 and S27 in FIG.

  Step S61 in FIG. 10 is the same as step S51 in FIG. 9, and waits in step S61 until environmental information that can be determined to be suitable for content presentation is obtained, and step S61 is performed when the appropriate environmental information is obtained. Proceed to S62.

  In step S67 in FIG. 10, it is determined whether or not the question list is empty (similar to step S57 in FIG. 9), that is, whether or not all questions are completed from the question list created in step S60. If the list is empty and the questions are completed, the process proceeds to step S69, and if not, the process returns to step S61.

  Step S69 in FIG. 10 updates the prediction model as appropriate using the training data newly obtained by the flow in FIG. 10, just like step S58 in FIG. Whether to perform the update, which time range of training data to use for the update, and the like are the same as described in step S58. When the operation phase is further continued with the new prediction model, the flow of FIG. 10 may be continued under the new prediction model.

  (Supplement 2) The content presentation system 10 can be provided for each user having a user terminal. When a part of the content presentation system 10 is realized on a server on the network, a database or the like is prepared for each user, and common processing is entrusted to a common server, and then processing for each user is realized.

  Here, if the content presentation system 10 is provided completely independently for each user, the data collection phase of FIG. 3 needs to be performed for each user. However, since it is assumed that users with similar attributes are in a similar state suitable for presenting content, the data collection phase should be implemented for each user up to such a case. It is not necessarily efficient.

  Therefore, if the data collection phase has not been completed for a certain user A, and there is another user B with similar attributes and the data collection phase has already been performed in the other user B, training for the other user B is performed. The data [B] may be adopted as training data for the user A, and the data collection phase for the user A may be omitted.

  Here, for user attributes, predetermined items such as age, gender, job type, etc. are registered for each user in a server or the like in the content presentation system 10, and the registered attributes are compared between users. Thus, it can be automatically determined whether or not the users A and B are similar. Similarity determination by attribute comparison can be automatically determined by obtaining a similarity degree or the like according to a predetermined rule such that each item is completely matched or partially matched.

  In addition, the attribute of the user for which the training data is borrowed by the similarity determination and the training data are not the attribute and training data as actual data, but the actual data is analyzed on the system administrator side, etc. For example, an artificially prepared one may be used. For example, it is possible to classify according to the personality of the user and prepare “serious” type attributes and training data and “quick” type attributes and training data.

  An embodiment in which the data collection phase is omitted can be implemented as follows.

  First, attribute information [Bi] and training data [Bi] for a plurality of users Bi (i = 1, 2,..., N) are stored in a predetermined database. As described above, the attribute information [Bi] and the training data [Bi] may be actually acquired or may be artificially prepared.

  Next, the creation unit 4 receives the attribute information [A] of the user from the user A, searches the database, identifies the user Bk whose attribute information [Bk] is most similar, and Training data [Bk] is adopted as training data. After the adoption, the training phase and the operation phase in the embodiment described above may be performed.

  (Supplement 3) In addition to the embodiment described with reference to FIGS. 5 and 6, creation of training data by the creation unit 4 can be various according to the content and characteristics of the content, and the user can present the content. Training data may be created by associating the environmental information with the evaluation result in such a way that whether the environmental information at the time of receiving is suitable for content presentation is reflected in the evaluation result.

  For example, in the embodiment of FIG. 6, as an additional process when reaching step S35, when an nth (n ≧ 3) answer is obtained for a certain content, the environment information when presented n−1 times Training data may be created by associating [n-1] with the nth answer evaluation result [n].

  In addition, when there are a plurality of types of content to be presented to a certain user, training data may be created in another embodiment for each content type. For example, as described above, for game content, training data may be created in the first embodiment of FIG. 5, and for learning content, training data may be created in the second embodiment of FIG. In such a case, each prediction model is constructed with training data for each content type, and when environmental information determined to be appropriate for each content type is obtained, the content of that type is presented. can do. Or you may make it construct | assemble one prediction model used in common with all types of content using the whole training data acquired in each embodiment.

  (Supplement 4) The present invention can also be provided as an operation method of the content presentation system 10, and also as a program (which may be stored in a storage medium) that causes a computer to function as the content presentation system 10. Can be provided.

  10 ... Content presentation system, 21 ... Sensor, 22 ... Environmental information acquisition unit, 31 ... Content database, 32 ... Management unit, 33 ... Presentation unit, 34 ... Reception unit, 4 ... Creation unit, 5 ... Construction unit

Claims (12)

An environment information acquisition unit for acquiring environment information of the user terminal;
A presentation unit for presenting content to the user;
A reception unit that receives a response to the presented content from the user;
Managing the content presented and receiving the response, and a management unit for obtaining an evaluation result for the received response;
A creation unit for creating training data by linking the acquired environmental information and the obtained evaluation result;
A content presentation comprising: a construction unit for constructing a prediction model as to whether or not the environmental information corresponds to an appropriate state for presenting the content to the user based on the training data system.   The construction unit constructs the prediction model on the basis of the training data on the assumption that environmental information when it can be determined that the evaluation result is good corresponds to an appropriate state for presenting content to the user. The content presentation system according to claim 1.   The construction unit uses the training data as learning data, and constructs a discriminant that outputs whether or not the environmental information may correspond to an appropriate state by machine learning as input by the environmental information, The content presentation system according to claim 2, wherein the prediction model is constructed. When the presentation unit presents content to the user and a response corresponding to the content is received by the reception unit, the environment information acquisition unit acquires environment information, and the management unit displays the evaluation result. Seeking
4. The content presentation system according to claim 1, wherein the creation unit creates training data by associating the environmental information acquired together with the obtained evaluation result. When it is determined that the evaluation result obtained by the management unit is bad when the presentation unit presents content to the user and the response corresponding to the content is received by the reception unit, the environment information acquisition unit Get environmental information,
The creating unit
The acquired environmental information and
When the content is presented to the user again by the presentation unit and the response is received by the reception unit, the evaluation result obtained by the management unit,
The content presentation system according to any one of claims 1 to 4, wherein training data is created by associating.   The content presentation system according to claim 1, wherein the content is content related to a problem for the user to learn, and the response is an answer by the user to the problem.   The management unit manages the content as a problem to be presented and received an answer in a question list, and for the problem determined that the evaluation result obtained for the answer received by the reception unit is good The content presentation system according to claim 6, wherein the content presentation system is excluded from subjects to be presented to the user by the presenting unit at a later time by being deleted from the question list.   The creation unit receives user attribute information from a user, and employs training data already created by another user with similar attribute information as training data of the user. 8. The content presentation system according to any one of items 7 to 7.   The environmental information acquisition unit receives at least one type of information selected from the group consisting of position information, date / time information, environmental sound information, image information, illuminance information, orientation information, acceleration information, temperature information, humidity information, and weather information. The content presentation system according to any one of claims 1 to 8, wherein environment information of the user terminal is acquired.   The presenting unit applies the prediction model constructed by the construction unit to the environmental information acquired by the environmental information acquisition unit, and determines that the environmental information corresponds to an appropriate state 10. The content presentation system according to claim 1, wherein the content is presented to a user. A step of periodically obtaining environmental information of the user terminal;
For the content managed in advance, periodically presenting the content to the user and receiving a response to the presented content from the user;
Creating training data by associating the acquired environmental information and the obtained evaluation result;
Constructing a prediction model based on the training data to construct whether or not the environmental information corresponds to an appropriate state for presenting the content to the user. .   A content presentation program for causing a computer to function as the content presentation system according to any one of claims 1 to 10.