JP7091700B2 - Information processing program, message analysis program, information processing device and information processing method - Google Patents

Description

The present invention relates to an information processing program, a message analysis program, an information processing apparatus, and an information processing method.

For example, the technique of an event estimation system that estimates information about a specific event by analyzing the content posted on Twitter or the like is widespread. In the event estimation system, the posted content can be analyzed, and for example, the disaster occurrence area can be specified by analyzing the information about the disaster.

Japanese Unexamined Patent Publication No. 2012-243032

In the event estimation system, for example, when analyzing the content posted on Twitter or the like, information on a specific event is extracted based on the keyword. However, in the event estimation system, in order to acquire information about a specific event with high accuracy, it is necessary to select an appropriate keyword for grasping the specific event.

One aspect is to provide an information processing program or the like that can select appropriate keywords for acquiring information on a specific event.

In one embodiment, the computer is made to execute a process of specifying the number of occurrences of a predetermined word included in a plurality of posts in a predetermined period. Further, when the number of appearances of the specified predetermined word is equal to or greater than the predetermined number of times, the computer is subjected to a process of extracting one or a plurality of words different from the predetermined word from the posts containing the predetermined word. Let it run. The one or more words are output to the computer depending on whether or not the number of appearances of each of the extracted one or more words is more than a specific threshold value as compared with the number of appearances in a period different from the predetermined period. To execute the processing to be performed.

One aspect is the ability to select appropriate keywords to obtain information about a particular event.

FIG. 1 is an explanatory diagram showing an example of the event estimation system of this embodiment. FIG. 2 is an explanatory diagram showing an example of the hardware configuration of the information processing apparatus. FIG. 3 is an explanatory diagram showing an example of the functional configuration of the information processing apparatus of the first embodiment. FIG. 4 is an explanatory diagram showing an example of processing when a child keyword is output from a tweet group including a parent keyword. FIG. 5 is an explanatory diagram showing an example of a candidate word storage unit. FIG. 6 is a flow chart showing an example of the processing operation of the control unit in the information processing apparatus related to the output processing. FIG. 7 is a flow chart showing an example of the processing operation of the control unit in the information processing apparatus related to the child keyword selection processing. FIG. 8 is an explanatory diagram showing an example of the functional configuration of the information processing apparatus of the second embodiment. FIG. 9 is an explanatory diagram showing an example of a word combination related to an event from a group of tweets including the keyword “earthquake” when an earthquake occurs in Kagoshima. FIG. 10 is an explanatory diagram showing an example of the target condition. FIG. 11 is an explanatory diagram showing an example of important words related to the current tweet group (tweet group when an earthquake occurred in Kagoshima) and the comparison target tweet group (tweet group related to past normal earthquakes). FIG. 12 is an explanatory diagram showing an example of the number of occurrences of each co-occurrence word related to an important word in a tweet group when an earthquake occurs in Kagoshima. FIG. 13 is an explanatory diagram showing an example of event estimation from a group of tweets during a traffic jam. FIG. 14 is an explanatory diagram showing an example of event estimation from a group of tweets during a traffic jam. FIG. 15 is an explanatory diagram showing an example of event estimation from a group of tweets when an earthquake occurs in Kagoshima. FIG. 16 is an explanatory diagram showing an example of event estimation from a group of tweets when soliciting earthquake relief funds. FIG. 17 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Kitami City, Hokkaido. FIG. 18 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Yokohama City, Kanagawa Prefecture. FIG. 19 is a flow chart showing an example of the processing operation of the control unit in the information processing apparatus related to the first estimation processing. FIG. 20 is a flow chart showing an example of the processing operation of the control unit in the information processing apparatus related to the first output processing. FIG. 21 is an explanatory diagram showing an example of the functional configuration of the information processing apparatus of the third embodiment. FIG. 22 is an explanatory diagram showing an example of the number of occurrences of each co-occurrence word related to an important word in a tweet group when an earthquake occurs in Kagoshima. FIG. 23 is an explanatory diagram showing an example of event estimation from a group of tweets when an earthquake occurs in Kagoshima. FIG. 24 is an explanatory diagram showing an example of event estimation from a group of tweets when soliciting earthquake relief funds. FIG. 25 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Kitami City, Hokkaido. FIG. 26 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Yokohama City, Kanagawa Prefecture. FIG. 27 is a flow chart showing an example of the processing operation of the control unit in the information processing apparatus related to the second estimation processing. FIG. 28 is a flow chart showing an example of the processing operation of the control unit in the information processing apparatus related to the second output processing. FIG. 29 is an explanatory diagram showing an example of an information processing apparatus that executes an information processing program.

Hereinafter, examples of the information processing program and the like disclosed in the present application will be described in detail with reference to the drawings. The disclosed technique is not limited by the present embodiment. In addition, the examples shown below may be appropriately combined as long as they do not cause a contradiction.

FIG. 1 is an explanatory diagram showing an example of the event estimation system 1 of this embodiment. The event estimation system 1 shown in FIG. 1 has a plurality of terminal devices 2, an information processing device 3, and a communication network 4. The event estimation system 1 is, for example, an SSS (Social Sensor System) that analyzes the content of a tweet by AI (Artificial Intelligence) technology and estimates, for example, a case where a disaster has occurred based on the analysis result. In the event estimation system 1, a location where a disaster has occurred may be estimated and displayed on a map. The plurality of terminal devices 2 are information terminals such as smartphones, tablet terminals, and personal computers that are connected to the communication network 4 and enjoy services such as SNS (Social Network Service) through the communication network 4. The information processing device 3 is, for example, a computer that connects to a communication network 4 and estimates an event from a keyword used in a tweet in an SNS. It should be noted that SNS is preferably a posting service such as Twitter that can read posts by an unspecified number of users, but is not limited to this, and is not limited to this, for example, a posting service such as Facebook that can read posts by a specific number of users. However, it can be changed as appropriate.

FIG. 2 is an explanatory diagram showing an example of the hardware configuration of the information processing apparatus 3. The information processing device 3 shown in FIG. 2 includes a communication device 11, an HDD (Hard Disk Drive) 12, a ROM (Read Only Memory) 13, a RAM (Random Access Memory) 14, and a CPU (Central Processing Unit) 15. , And a bus 16. The communication device 11 is a device that communicates and connects with the communication network 4. The HDD 12 is a storage area for storing various information. The ROM 13 is a storage area for storing information such as various programs. The RAM 14 is a storage area for storing various types of information. The CPU 15 controls the entire information processing apparatus 3. The bus 16 is a communication line, a control line, or the like that transmits / receives data to / from the communication device 11, the HDD 12, the ROM 13, the RAM 14, and the CPU 15.

FIG. 3 is an explanatory diagram showing an example of the functional configuration of the information processing apparatus 3. The information processing device 3 shown in FIG. 3 has a control unit 3A and a storage unit 3B. The control unit 3A corresponds to, for example, the CPU 15. The control unit 3A expands the information processing program stored in the ROM 13 on the RAM 14, and executes the information processing program expanded on the RAM 14 as an information processing process. Therefore, for example, the specific unit 21 and the extraction unit 22, the comparison unit 23, the output unit 24, and the estimation unit 25 are executed as functions. The storage unit 3B corresponds to, for example, HDD 12, ROM 13, RAM 14, and the like. The storage unit 3B has a keyword storage unit 31 and a candidate word storage unit 32. The keyword storage unit 31 is an area for storing keywords such as a parent keyword and a child keyword. The parent keyword is a word for extracting a tweet expressing a specific event. The child keyword is a word that assists in extracting tweets expressing a specific event in addition to the parent keyword. The candidate word storage unit 32 is, for example, an area for storing candidate words. The candidate word is a child keyword candidate word.

The specific unit 21 counts the number of tweets including the parent keyword per unit time, and determines that a burst has occurred when the number of tweets reaches t times the number of tweets in normal times. The extraction unit 22 performs morphological analysis on the document of the tweet group including the parent keyword during the burst period, and divides the tweet group into words. After the word is divided, the extraction unit 22 ranks the words that frequently appear in the tweet group according to the number of occurrences. Further, the extraction unit 22 extracts words having the highest number of occurrences as candidate words, and stores the extracted candidate words in the candidate word storage unit 32. The extraction unit 22 extracts the first candidate word, which is the candidate word for the current burst period, and the second candidate word, which is the candidate word for the burst period for the past X times, from the candidate word storage unit 32.

The comparison unit 23 compares the first candidate word with the second candidate word. The comparison unit 23 determines whether or not the number of candidate words of the same word among the first candidate word and the second candidate word is a predetermined number M or more. When the number of candidate words of the same word is a predetermined number M or more, the comparison unit 23 selects the candidate words of the predetermined number M or more as child keywords.

The output unit 24 associates the selected child keyword with the parent keyword and stores it in the keyword storage unit 32. The estimation unit 25 estimates the content of the case from the tweet group using the parent keyword stored in the keyword storage unit 32 and the child keyword corresponding to the parent keyword.

FIG. 4 is an explanatory diagram showing an example of processing when a child keyword is output from a tweet group including a parent keyword. When the information processing device 3 detects a burst of tweets including the parent keyword "power failure", the tweet group during the current burst period including the parent keyword "power failure" and the past X times including the parent keyword "power failure" Compare with the group of tweets during the burst period. Based on the comparison result, the information processing apparatus 3 determines that the first candidate word "breaker" and the second candidate word "breaker" are candidate words of a predetermined number of M or more, and uses "breaker" as a child keyword. Is output as.

FIG. 5 is an explanatory diagram showing an example of the candidate word storage unit 32. The candidate word storage unit 32 shown in FIG. 5 manages each burst period of the tweet group including the parent keyword, and manages the candidate word and the number of appearances of the candidate word in association with the ascending order of the number of appearances. The candidate word has a first candidate word and a second candidate word, the first candidate word is the candidate word during the current burst period, and the second candidate word is the past. It is a candidate word during the burst period of X times. The control unit 3A refers to the candidate word storage unit 32, and the number of appearances of the candidate word and the candidate word during the burst period including the parent keyword, and the number of appearances of the candidate word and the candidate word during the past burst period including the parent keyword. Can be recognized. The comparison unit 23 refers to the candidate word storage unit 32 shown in FIG. 5, and selects, for example, “OK”, “seismic intensity”, and “power failure” as child keywords.

FIG. 6 is a flow chart showing an example of the processing operation of the control unit 3A in the information processing apparatus 3 related to the output processing. In FIG. 6, the specific unit 21 in the control unit 3A collects tweets including the parent keyword from the tweet group posted in the SNS at regular time intervals (step S11).

The specific unit 21 counts the number of tweets including the parent keyword acquired within the unit time (step S12). The specific unit 21 determines whether or not the number of tweets including the parent keyword exceeds a predetermined number of times (step S13). The predetermined number of times is, for example, t times the number of tweets in normal times. When the number of tweets including the parent keyword exceeds a predetermined number of times (step S13 affirmative), the specific unit 21 determines that a burst of tweets including the parent keyword has occurred (step S14). After determining that a burst has occurred, the control unit 3A executes the child keyword selection process shown in FIG. 7, which will be described later (step S15). The child keyword is a keyword for appropriately estimating the event of the parent keyword. The output unit 24 in the control unit 3A outputs the child keyword selected in the child keyword selection process (step S16), and ends the process operation shown in FIG. The control unit 3A stores the output child keyword in the keyword storage unit 31. When the number of tweets does not exceed the predetermined number of times (step S13 denial), the specific unit 21 proceeds to step S11 in order to acquire the tweets including the parent keyword.

As a result, the estimation unit 25 in the control unit 3A can estimate the content of the case from the tweet group with high accuracy by using the parent keyword stored in the keyword storage unit 31 and the child keyword corresponding to the parent keyword.

FIG. 7 is a flow chart showing an example of the processing operation of the control unit 3A in the information processing apparatus 3 related to the child keyword selection processing. In FIG. 7, the extraction unit 22 in the control unit 3A analyzes the document including the tweet group of the parent keyword during the burst period by morphological analysis (step S21), and divides the document after the morphological analysis into words (step S22).

The extraction unit 22 counts the number of occurrences of each word (step S23), and stores the words having the highest number of appearances up to the Nth place in the candidate word storage unit 32 as the current candidate word (step S24). In addition to the candidate words during the burst period, the candidate word storage unit 32 stores the candidate words stored in the past and having the highest number of appearances. The extraction unit 22 extracts the candidate word in the current burst period as the first candidate word from the candidate word storage unit 32 (step S25). Further, the extraction unit 22 extracts the candidate words for the past X times as the second candidate word from the candidate word storage unit 32 (step S26).

The comparison unit 23 in the control unit 3A compares the first candidate word with the second candidate word (step S27). The comparison unit 23 determines whether or not there are a predetermined number or more of the candidate words of the same word among the first candidate word and the second candidate word (step S28). When the number of candidate words of the same word is equal to or more than a predetermined number (step S28 affirmative), the comparison unit 23 selects the predetermined number or more of candidate words as child keywords (step S29), and ends the processing operation shown in FIG. 7. When the number of candidate words of the same word is not equal to or more than a predetermined number (denial in step S28), the control unit 3A ends the processing operation shown in FIG. 7.

The control unit 3A compares the candidate words up to the top N in the burst period of this time with the candidate words up to the top N in the burst period of the past X times, so that it can respond to changes in words, for example. Keyword extraction can be realized. Note that N, X and M can be adjusted according to the event and operation.

The information processing apparatus 3 of the first embodiment can select an appropriate child keyword for estimating an event. Further, in the information processing apparatus 3, since the candidate word having a higher number of appearances is likely to be a word representing the event, a more appropriate child keyword can be selected by comparing the past X cases. Furthermore, the information processing device 3 can respond to changes with the times and the wording peculiar to Twitter by comparing cases X times in the past. Moreover, since the event is estimated using appropriate keywords, the accuracy of event estimation is improved.

In the information processing apparatus 3 of the first embodiment, when it is detected that a burst of a tweet group including a parent keyword has occurred, a candidate word is extracted from the tweet group including the parent keyword. The information processing apparatus 3 outputs the candidate word as a child keyword depending on whether or not the number of appearances of each candidate word is a predetermined number M or more as compared with the number of appearances of the candidate word in the burst period for the past X times. As a result, the candidate word with the highest number of occurrences in the past X case comparisons is selected as an appropriate child keyword for estimating the event, and the predictability of the event is improved.

The comparison unit 23 of the first embodiment uses the number of appearances of the candidate word in the burst period of the past X times as a comparative example, but is not limited to the burst period of the past X times, and the number of appearances is a predetermined number. It may be a period of less than the past X bursts not detected, and can be changed as appropriate.

The comparison unit 23 of the first embodiment selects the same candidate word having a higher number of appearances as a child keyword according to the number of appearances of the first candidate word and the number of appearances of the second candidate word. However, in the comparison unit 23, the first ratio of the number of appearances of the candidate word in the current burst period to the number of appearances of the candidate word in the tweet group for the past X times is the number of appearances of the parent keyword in the tweet group for the past X times. This time, it is determined whether or not it matches or is similar to the second rate of the number of occurrences of the parent keyword during the burst period. When the first rate matches or is similar to the second rate, the comparison unit 23 may output the candidate word as a child keyword and can change it as appropriate.

The comparison unit 23 of the first embodiment selects the same candidate word having a higher number of appearances as a child keyword according to the number of appearances of the first candidate word and the number of appearances of the second candidate word. However, child keywords may be selected according to the number of occurrences of the first candidate word and the second candidate word.

The comparison unit 23 may output the candidate word according to the number of appearances of the candidate word, or may output the candidate word according to the increasing tendency of the number of appearances of the candidate word, and can be changed as appropriate.

The specific unit 21 of the first embodiment determines that a burst of tweets including the parent keyword occurs when the number of tweets including the parent keyword exceeds a predetermined number of times t times the number of tweets in normal times. However, the number of tweets including the parent keyword is not limited to the number of occurrences of the parent keyword, and can be changed as appropriate.

In the event estimation system of the first embodiment, a specific event was estimated from the tweet group using the parent keyword and the child keyword. However, important words and co-occurrence words may be extracted from a group of tweets including a parent keyword and a child keyword, and a specific event may be estimated using the important words and co-occurrence words. This will be described below. The same configuration as that of the event estimation system 1 of the first embodiment is designated by the same reference numeral, and the description of the overlapping configuration and operation will be omitted.

FIG. 8 is an explanatory diagram showing an example of the functional configuration of the information processing apparatus 3 of the second embodiment. The information processing device 3 shown in FIG. 8 has a control unit 3C and a storage unit 3D. The control unit 3C corresponds to, for example, the CPU 15. The control unit 3C, for example, expands the information processing program stored in the ROM 13 on the RAM 14 and executes the information processing program expanded on the RAM 14 as an information processing process. 42, the important word identification unit 43, the co-occurrence word identification unit 44, the output unit 45, and the estimation unit 46 are executed as functions. The storage unit 3D corresponds to, for example, HDD 12, ROM 13, RAM 14, and the like.

The storage unit 3D has a keyword storage unit 51, an important word storage unit 52, and a co-occurrence word storage unit 53. The keyword storage unit 51 is an area for storing keywords such as a parent keyword and a child keyword, for example. The important word storage unit 52 is, for example, an area for storing important words. The co-occurrence word storage unit 53 is, for example, an area for storing co-occurrence words.

The specific unit 41 collects tweets at regular intervals under certain conditions such as keywords. The specific unit 41 detects the occurrence of a burst of tweets under certain conditions. The burst occurrence of a tweet group under a certain condition is, for example, a temporary increase in message transmission including a predetermined word. The extraction unit 42 acquires the first document including the current tweet group during the burst period. Further, the extraction unit 42 acquires a second document including a comparison target tweet group that matches the target condition from the past normal tweet group. The target condition is, for example, a condition for extracting a comparison target tweet group from a tweet group in units of day, time, day of the week, time zone, or number of tweets. In addition, the target condition is a condition for extracting a comparison target tweet group during the period inside and outside the burst period. The daily extraction condition is suitable when comparing with the tweet group of the day when the burst is detected, the tweet group for the past D days is targeted for comparison, and for example, when the event such as a concert event is distinguished on a daily basis. .. The time-based extraction condition is when comparing with the past tweet group of time H when the burst is detected, the tweet group for the past D × H hours is targeted for comparison, and for example, when an event such as a traffic jam is distinguished by time unit. Suitable for. The extraction condition for each day of the week is suitable for comparing past tweets on the day when the burst is detected, for example, for distinguishing an event such as a train delay that is greatly affected by the day of the week characteristics. The extraction condition for each time zone is suitable for comparing past tweet groups in the time zone in which a burst is detected, and for distinguishing events that are greatly affected by the time zone, such as commuting status during a typhoon. The extraction condition for each number of tweets is to compare the same number of past tweets as the number of tweets while the burst was detected. For example, it is not a general event such as a water pipe rupture, and it is not posted frequently in normal times. , Suitable for distinguishing events where a sufficient number of tweets cannot be obtained.

The important word identification unit 43 morphologically analyzes the first document including the current tweet group and the second document including the comparison target tweet group, and divides the words into words. The important word identification unit 43 calculates the TF-IDF value for each word using (Equation 1), (Equation 2) and (Equation 3). The important word identification unit 43 stores a word having a higher TF-IDF value as an important word in the important word storage unit 52.

In addition, the correction value of the word type used in (Equation 3) is an important word, but when the IDF value is low, for example, although the number is small from usual, it appears itself and is large at the time of burst. This is a correction value when applied to the words that appear. The correction value may be adjusted according to the information to be known such as a place, a proper noun, a verb, and an adjective (state). For example, if you want to know the place, proper noun and state, set "place" to 1, "proper noun" to 1, "verb" to 0, and "adjective" to 0.5 as correction values. That is, as the correction value increases, the TF-IDF value becomes higher.

In the first document including the current tweet group, the co-occurrence word identification unit 44 becomes an important word representing the event as the TF-IDF value becomes higher. The co-occurrence word identification unit 44 identifies a word having a higher TF-IDF value for the first document in units of one tweet, that is, a co-occurrence word that co-occurs with an important word, and co-occurrence of the co-occurrence word of the important word. Count the number of times. The co-occurrence word identification unit 44 stores the co-occurrence words having the higher number of co-occurrence of important words in the co-occurrence word storage unit 53.

The output unit 45 outputs to the estimation unit 46 a combination of the important word stored in the important word storage unit 52 and the co-occurrence word corresponding to the important word stored in the co-occurrence word storage unit 53. The estimation unit 46 estimates the case according to the combination of the important word and the co-occurrence word.

FIG. 9 is an explanatory diagram showing an example of a word combination related to an event from a group of tweets including the keyword “earthquake” when an earthquake occurs in Kagoshima. The important word identification unit 43 identifies the important word “Kagoshima” at the top of TF-IDF from the tweet group including “earthquake”. Furthermore, the co-occurrence word identification unit 44 has the highest number of appearances among the words that co-occur to the important word "Kagoshima" from the current tweet group including "earthquake" and the past tweet group including "earthquake". Identify the words "seismic intensity 5+", "large" and "earthquake source". Then, the estimation unit 46 estimates the event that "an earthquake with a seismic intensity of 5 or higher occurred in Kagoshima" by combining the important word "Kagoshima", the co-occurrence words "seismic intensity 5 upper", "large" and "earthquake source". do.

FIG. 10 is an explanatory diagram showing an example of the target condition. The extraction unit 42 in the control unit 3C extracts a comparison target tweet group that matches the target condition for each time zone from the past tweet group. The target condition shown in FIG. 10 is an extraction condition for each time zone. When the tweet group of this burst period is in the time zone from 10:00 to 18:00 on August 3 and the target condition is in the time zone unit, for example, the past August 2nd and August 1st , July 31 ... The tweet group in the time zone from 10:00 to 18:00 is extracted from the past tweet group as the comparison target tweet group.

FIG. 11 is an explanatory diagram showing an example of important words related to the current tweet group (tweet group when an earthquake occurred in Kagoshima) and the comparison target tweet group (tweet group related to past normal earthquakes). Currently, the tweet group is the tweet group during this burst period, and is the tweet group when the earthquake occurred in Kagoshima. The comparison target tweet group is a tweet group related to past normal earthquakes extracted by the extraction unit 42. The important word identification unit 43 in the control unit 3C analyzes the first document of the tweet group of this time and the second document of the tweet group to be compared by morphological analysis and divides the words. Further, the important word specifying unit 43 calculates the TF-IDF value for each divided word, and extracts the word having the higher TF-IDF value as the important word.

As shown in FIG. 11, the important word identification unit 43 ranks first as "Kagoshima", second as "power outage", and third as important words of the upper TF-IDF of the current tweet group and the comparison target tweet group. The rank is ranked as "landslide" and stored in the important word storage unit 52. As for the important words shown in FIG. 11, there is a tendency that a place or a word peculiar to the occurrence of an earthquake is more likely to appear at the upper level of TF-IDF than a word commonly appearing at the time of an earthquake.

FIG. 12 is an explanatory diagram showing an example of the number of occurrences of each co-occurrence word related to an important word in a tweet group when an earthquake occurs in Kagoshima. The co-occurrence word identification unit 44 in the control unit 3C identifies the co-occurrence word of the important word from the current tweet group this time. Then, the co-occurrence word identification unit 44 counts the number of co-occurrence words, which is the number of appearances of the co-occurrence words for each important word. As shown in FIG. 12, the co-occurrence word identification unit 44 has, for example, two co-occurrence words of "earthquake source" and two co-occurrence words of "landslide" as co-occurrence words of the important word "Kagoshima". , The number of co-occurrence of "power failure" is set to 1 and stored in the co-occurrence word storage unit 53.

FIG. 13 is an explanatory diagram showing an example of event estimation from a group of tweets during a traffic jam. It is assumed that the important word identification unit 43 identifies the important words "Kiyama", "Kameyama" and "Ichinomiya" with the keyword "traffic jam". Further, in the co-occurrence word identification unit 44, the co-occurrence word of the important word "Kiyama" is "near", "up", "down" and "accident", and the co-occurrence word of the important word "Kameyama" is "near". It is assumed that the co-occurrence words of "IC", "up" and "down", and the important word "Ichinomiya" are specified as "nearby", "IC" and "up". From the combination of these important words and co-occurrence words, the estimation unit 46 can estimate the case of "up and down traffic jams at Kiyama and Kameyama, up and down traffic jams at Ichinomiya, and an accident".

FIG. 14 is an explanatory diagram showing an example of event estimation from a group of tweets during a traffic jam. It is assumed that the important word identification unit 43 specifies that the keyword is "traffic jam" and the important words are "aqua line", "Kisarazu" and "return home". Further, in the co-occurrence word identification unit 44, the co-occurrence words of the important word "Aqualine" are "time", "up" and "return", and the co-occurrence words of the important word "Kisaratsu" are "up" and "IC". And, it is assumed that the co-occurrence word of "nearby" and the important word "homecoming" is specified as "time". From the combination of these important words and co-occurrence words, the estimation unit 46 can estimate a case of "traffic jam occurs at the time of returning home near the aqua line and Kisarazu".

FIG. 15 is an explanatory diagram showing an example of event estimation from a group of tweets when an earthquake occurs in Kagoshima. It is assumed that the important word identification unit 43 identifies the keywords as "earthquake" and the important words as "Kagoshima", "power outage", "earthquake" and "rain". Furthermore, in the co-occurrence word identification unit 44, the co-occurrence words of the important word "Kagoshima" are "seismic intensity", "Kyushu", "OK", "worry", "power failure" and "earthquake source", and the important word "power failure". Suppose that the co-occurrence words are specified as "Kagoshima", "Kyushu", and "South". Further, in the co-occurrence word identification unit 44, the co-occurrence words of the important word "landslide" are "Kyushu", "Kagoshima" and "worry", and the co-occurrence words of the important word "rain" are "Kyushu", "OK". Suppose that you have identified "Kagoshima" and "worry". From the combination of these important words and co-occurrence words, the estimation unit 46 can estimate a case of "an earthquake with Kagoshima as the epicenter, a power outage, and a concern about landslides due to rain."

FIG. 16 is an explanatory diagram showing an example of event estimation from a group of tweets when soliciting earthquake relief funds. It is assumed that the important word identification unit 43 identifies the keywords as "earthquake" and the important words as "suma", "support", "donation" and "support". Furthermore, in the co-occurrence word identification unit 44, the co-occurrence words of the important word "suma" are "reconstruction", "earthquake", "eastern Japan", "Kumamoto" and "great earthquake", and the important word "support" is "reconstruction". , "Eastern Japan", "Great Earthquake", "Kumamoto" and "Earthquake". Furthermore, the co-occurrence words of the important word "donation" are "reconstruction", "Kumamoto", "earthquake", "eastern Japan" and "request", and the co-occurrence words of the important word "support" are "reconstruction", "great earthquake", It is assumed that "Eastern Japan", "Kumamoto" and "earthquake" are identified. From the combination of these important words and co-occurrence words, the estimation unit 46 can estimate the case of "the earthquake did not occur, but the support for the earthquake in Kumamoto and eastern Japan".

FIG. 17 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Kitami City, Hokkaido. It is assumed that the important word identification unit 43 specifies that the keyword is "power failure" and the important words are "thunder", "Kitami", "moment" and "signal". Further, in the co-occurrence word identification unit 44, the co-occurrence words of the important word "thunder" are "laughs" and "Kitami", and the co-occurrence words of the important words "Kitami" are "city", "information" and "laughs". , The co-occurrence words of the important word "moment" are identified as "thunder", "laugh" and "electricity", and the co-occurrence words of the important word "signal" are identified as "electricity", "middle", "house" and "revival". Suppose you did. From the combination of these important words and co-occurrence words, the estimation unit 46 can estimate the case that "a power outage caused by lightning occurred in Kitami City, Hokkaido, and the signal and the like were affected by the momentary power outage."

FIG. 18 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Yokohama City, Kanagawa Prefecture. It is assumed that the important word identification unit 43 specifies that the keyword is "power outage" and the important words are "middle", "Yokohama" and "uchi". Further, in the co-occurrence word identification unit 44, the co-occurrence words of the important word "middle" are "city", "ward" and "Yokohama", and the co-occurrence words of the important word "Yokohama" are "middle", "city" and It is assumed that the co-occurrence words of "ward" and the important word "uchi" are specified as "neighborhood" and "area". The estimation unit 46 can estimate the case of "a power outage has occurred in Naka-ku, Yokohama" from the combination of important words and co-occurrence words.

FIG. 19 is a flow chart showing an example of the processing operation of the control unit 3C in the information processing apparatus 3 related to the first estimation processing. In FIG. 19, the specific unit 41 in the control unit 3C collects tweets with certain conditions from the tweets posted on the SNS (step S31). The specific unit 41 counts the number of tweets under a certain condition within a unit time (step S32). The specific unit 41 determines whether or not the number of tweets under a certain condition exceeds a predetermined number of times (step S33). The predetermined number of times is, for example, t times the number of tweets in normal times. When the number of tweets under a certain condition exceeds a predetermined number of times (step S33 affirmative), the specific unit 41 determines that the number of message transmissions is temporarily increased, that is, the occurrence of vertos of the tweet group under a certain condition (step S34). The control unit 3C executes the first output process shown in FIG. 20 (step S35).

After executing the first output process, the estimation unit 46 in the control unit 3C estimates an event according to the combination of the important word and the co-occurrence word (step S36), and ends the processing operation shown in FIG. .. When the number of tweets under a certain condition does not exceed the predetermined number of times (step S33 is denied), the estimation unit 46 proceeds to step S31 in order to acquire the tweets under a certain condition.

FIG. 20 is a flow chart showing an example of the processing operation of the control unit 3C in the information processing apparatus 3 related to the first output processing. In FIG. 20, the extraction unit 42 in the control unit 3C acquires the first document including the current tweet group during the burst period (step S41). The extraction unit 42 acquires a second document including a comparison target tweet group that matches the target condition from the past normal tweet group (step S42).

The important word identification unit 43 in the control unit 3C divides words by morphological analysis of the first document and the second document (step S43), and calculates a TF-IDF value for each word (step S44). The important word specifying unit 43 identifies words having a TF-IDF value up to the highest N rank for each word as important words (step S45). The important word specifying unit 43 stores the specified important word in the important word storage unit 52.

The co-occurrence word specifying unit 44 in the control unit 3C identifies a co-occurrence word for each important word as a co-occurrence word, and counts the number of co-occurrence words for each co-occurrence word (step S46). The co-occurrence word is, for example, a noun word co-occurring with an important word. The co-occurrence word specifying unit 44 stores the specified co-occurrence word in the co-occurrence word storage unit 53.

The output unit 45 in the control unit 3C outputs the co-occurrence words up to the top N in ascending order of the number of co-occurrence words among the co-occurrence words of the important words in combination with the important words (step S47), and is shown in FIG. End the processing operation.

The information processing apparatus 3 of the second embodiment identifies an important word peculiar to an event, and also identifies a co-occurrence word that appears particularly prominently when the event occurs, as compared with other comparison target tweet groups acquired under the same conditions. do. The information processing apparatus 3 estimates an event according to a combination with a co-occurrence word that is often muttered at the same time as an important word. As a result, it is possible to grasp the main points of the phenomenon and the cause peculiar to the event.

When the information processing apparatus 3 detects a temporary increase in message transmission including important words, the information processing apparatus 3 extracts co-occurrence words that co-occur with the important words for the messages transmitted in the period corresponding to the temporary increase. The information processing apparatus 3 outputs a combination of important words and co-occurrence words, and estimates an event according to these important words and co-occurrence words. As a result, it is possible to grasp the main points of the phenomenon and the cause peculiar to the event.

In the information processing apparatus 3 of the second embodiment, the co-occurrence words co-occurring for each important word are exemplified by nouns, but the present invention is not limited to this, and for example, even if the co-occurrence words are adjectives. Well, the embodiment thereof will be described below as Example 3.

FIG. 21 is an explanatory diagram showing an example of the functional configuration of the information processing apparatus 3 of the third embodiment. The same configuration as that of the information processing apparatus 3 of the second embodiment is designated by the same reference numeral, and the description of the overlapping configuration and operation will be omitted.

The information processing device 3 shown in FIG. 21 has a control unit 3E and a storage unit 3F. The control unit 3E corresponds to, for example, the CPU 15. The control unit 3E expands the information processing program stored in the ROM 13 on the RAM 14, and executes the information processing program expanded on the RAM 14 as an information processing process. 42, the important word identification unit 43, the co-occurrence word identification unit 44A, the output unit 45, and the estimation unit 46 are executed as functions. The storage unit 3F corresponds to, for example, HDD 12, ROM 13, RAM 14, and the like.

The storage unit 3F has a keyword storage unit 51, an important word storage unit 52, and a co-occurrence word storage unit 53. The keyword storage unit 51 is an area for storing keywords such as a parent keyword and a child keyword, for example. The important word storage unit 52 is, for example, an area for storing important words. The co-occurrence word storage unit 53 is, for example, an area for storing co-occurrence words of adjectives.

The co-occurrence word specifying unit 44A identifies a co-occurrence word of an adjective among the words co-occurring with an important word. The co-occurrence word specifying unit 44 stores the co-occurrence word having the higher number of co-occurrence among the co-occurrence words of the adjective in the co-occurrence word storage unit 53.

FIG. 22 is an explanatory diagram showing an example of the number of occurrences of each co-occurrence word related to an important word in a tweet group when an earthquake occurs in Kagoshima. The co-occurrence word identification unit 44A identifies the co-occurrence word of the adjective of the important word from the current tweet group this time. Then, the co-occurrence word specifying unit 44A counts the number of co-occurrence words for each important word. As shown in FIG. 22, the co-occurrence word identification unit 44A has, for example, as a co-occurrence word of the important word "Kagoshima", the number of co-occurrence of "large" is 2 times and the number of co-occurrence of "scary" is 2 times. The number of co-occurrence of "many" is set to 1 and stored in the co-occurrence word storage unit 53.

FIG. 23 is an explanatory diagram showing an example of event estimation from a group of tweets when an earthquake occurs in Kagoshima. It is assumed that the important word identification unit 43 specifies that the keyword is "earthquake" and the important words are "Kagoshima", "recently" and "power outage". Further, in the co-occurrence word identification unit 44A, the co-occurrence words of the important word "Kagoshima" are "large", "many", "strong" and "scary", and the co-occurrence words of the important word "recent" are "many". Suppose that the co-occurrence words "scary" and "big" and the important word "power failure" are identified as "hot". From the combination of these important words and co-occurrence words, the estimation unit 46 can estimate the case that "a large earthquake has recently occurred frequently in Kagoshima and it has become hot due to a power outage."

FIG. 24 is an explanatory diagram showing an example of event estimation from a group of tweets when soliciting earthquake relief funds. It is assumed that the important word identification unit 43 identifies the keywords as "earthquake" and the important words as "reconstruction", "Kumamoto", "eastern Japan" and "support". Further, in the co-occurrence word identification unit 44A, the co-occurrence words of the important word "reconstruction" are "nice", "warm" and "deep", and the co-occurrence words of the important word "Kumamoto" are "nice", "warm" and The co-occurrence words of "warm" and the important word "East Japan" were identified as "nice", "warm" and "details", and the co-occurrence words of the important word "support" were identified as "nice", "warm" and "warm". And. From the combination of important words and co-occurrence words, the estimation unit 46 can estimate the case that "the earthquake did not occur and we are requesting warm support for the earthquakes in Kumamoto and eastern Japan."

FIG. 25 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Kitami City, Hokkaido. It is assumed that the important word identification unit 43 specifies that the keyword is "power outage" and the important words are "thunderstorm", "rain" and "Kitami". Further, in the co-occurrence word identification unit 44A, the co-occurrence words of the important word "thunder" are "wow", "bad" and "scary", and the co-occurrence words of the important word "rain" are "wow", "bad" and Suppose that the co-occurrence words of "scary" and the important word "Kitami" are identified as "wide", "dangerous", and "wide". From the combination of important words and co-occurrence words, the estimation unit 46 can estimate the case that "a power outage has occurred due to lightning, it is raining, and a power outage has occurred in a wide area of Kitami City."

FIG. 26 is an explanatory diagram showing an example of event estimation from a group of tweets when a power failure occurs in Yokohama City, Kanagawa Prefecture. It is assumed that the important word identification unit 43 identifies the keyword as "power outage" and the important words as "Yokohama" and "information". Further, it is assumed that the co-occurrence word identification unit 44A identifies the co-occurrence word of the important word "Yokohama" as "rare" and "horror", and the co-occurrence word of the important word "information" as "none". The estimation unit 46 estimates a case where "a power outage occurs rarely in Yokohama City and there is no information" from the combination of important words and co-occurrence words.

FIG. 27 is a flow chart showing an example of the processing operation of the control unit 3E in the information processing apparatus 3 related to the second estimation processing. In FIG. 27, the control unit 3E executes the second output process shown in FIG. 28 after determining in step S34 that the tweet group of tweets under certain conditions has been generated (step S35A). After executing the second output process, the estimation unit 46 estimates an event according to the combination of the important word and the co-occurrence word (adjective) (step S36A), and ends the process operation shown in FIG. 27. The co-occurrence word is, for example, an adjective.

FIG. 28 is a flow chart showing an example of the processing operation of the control unit 3E in the information processing apparatus 3 related to the second output processing. In FIG. 28, the important word identification unit 43 in the control unit 3E extracts a word having a higher TF-IDF value from the TF-IDF for each word as an important word in step S45. The co-occurrence word identification unit 44A in the control unit 3E counts the number of co-occurrence words of the adjectives that co-occur for each important word (step S46A). The co-occurrence word is, for example, an adjective.

The output unit 45 outputs the co-occurrence words of the adjectives up to the top N in ascending order of the number of co-occurrence words among the co-occurrence words of the important words in combination with the important words (step S47A), and performs the processing operation shown in FIG. 28. finish.

The information processing apparatus 3 identifies an important word peculiar to an event, and also identifies a co-occurrence word of an adjective that appears particularly prominently when the event occurs, as compared with other comparison target tweets acquired under the same conditions. The information processing apparatus 3 estimates an event according to a combination with a co-occurrence word of an adjective that is often muttered at the same time as an important word. As a result, it is possible to grasp the main points of the phenomenon and the cause peculiar to the event.

When the information processing device 3 detects a temporary increase in the transmission of tweets containing important words, it identifies a co-occurrence word of an adjective that co-occurs with the important word for the tweet transmitted during the period corresponding to the temporary increase. do. The information processing device 3 evaluates the period based on the transmission status of the tweet inside and outside the period including the co-occurrence word of the specified adjective. As a result, the period of this tweet group can be evaluated according to the co-occurrence words of important words and adjectives. You can register changes in words according to the adjectives that frequently appear for each event.

Further, each component of each of the illustrated parts does not necessarily have to be physically configured as shown in the figure. That is, the specific form of distribution / integration of each part is not limited to the one shown in the figure, and all or part of them may be functionally or physically distributed / integrated in any unit according to various loads and usage conditions. Can be configured.

Further, various processing functions performed by each device are performed on a CPU (Central Processing Unit) (or a microcomputer such as an MPU (Micro Processing Unit) or an MCU (Micro Controller Unit)) in whole or in any part thereof. You may try to do it. Further, various processing functions may be executed in whole or in any part on a program analyzed and executed by a CPU (or a microcomputer such as an MPU or MCU) or on hardware by wired logic. Needless to say.

By the way, various processes described in this embodiment can be realized by executing a program prepared in advance in the information processing apparatus. Therefore, in the following, an example of an information processing apparatus that executes a program having the same function as that of the above embodiment will be described. FIG. 29 is an explanatory diagram showing an example of an information processing apparatus that executes an information processing program.

The information processing device 100 that executes the information processing program shown in FIG. 29 includes a communication device 110, a display device 120, an operation device 130, an HDD 140, a ROM 150, a RAM 160, a CPU 170, and a bus 180. The communication device 110 connects to a communication network (not shown) and collects SNS posts in the communication network.

The ROM 150 stores in advance an information processing program that exhibits the same functions as those in the above embodiment. The information processing program does not necessarily have to be stored in the ROM 150 from the beginning, and the information processing program may be recorded on a recording medium that can be read by a drive (not shown). The recording medium may be, for example, a flexible disk (FD), a CD-ROM, a DVD disk, a USB memory, a portable recording medium such as an SD card or an IC card, a semiconductor memory such as a flash memory, or the like. The information processing apparatus 100 may read and execute the information processing program stored in the recording medium. Further, as shown in FIG. 29, the information processing program includes a specific program 150A, an extraction program 150B, a comparison program 150C, and an output program 150D. The programs 150A to 150D may be integrated or dispersed as appropriate.

Then, the CPU 170 reads these programs 150A to 150D from the ROM 150, and expands each of these read programs on the work area of the RAM 160. Then, the RAM 160 functions as each of the expanded programs 150A to 150D as a specific process 160A, an extraction process 160B, a comparison process 160C, and an output process 160D.

The CPU 170 specifies the number of occurrences of a predetermined word included in a plurality of posts in a predetermined period. When the number of appearances of the specified predetermined word is equal to or greater than the predetermined number of times, the CPU 170 extracts one or a plurality of words different from the predetermined word from the posts containing the predetermined word. The CPU 170 outputs one or a plurality of words depending on whether or not the number of appearances of each of the extracted one or a plurality of words is greater than or equal to a specific threshold value as compared with the number of appearances in a period different from a predetermined period. As a result, it is possible to select appropriate keywords for acquiring information on a specific event.

1 Event estimation system 3 Information processing device 21 Specific unit 22 Extraction unit 23 Comparison unit 24 Output unit 41 Specific unit 42 Extraction unit 43 Important word specific unit 44 Co-occurrence word specific unit 44A Co-occurrence word specific unit 45 Output unit

Claims (10)

Identify the number of occurrences of a given word for a particular event in multiple posts over a given time period
When the number of appearances of the specified predetermined word is equal to or greater than the predetermined number of times, the post containing the predetermined word is different from the predetermined word and co-occurs the predetermined word, 1 or Extract multiple co -occurrence words and
Depending on whether or not the number of appearances of each of the extracted one or more co -occurrence words is greater than or equal to a specific threshold value as compared with the number of appearances in a period different from the predetermined period, the one or more co -occurrence words may be selected. Output and
The specific event is estimated using the predetermined word and the output one or more co-occurrence words.
An information processing program characterized by having a computer execute processing. The information processing program according to claim 1, wherein the period different from the predetermined period is a period in which the number of appearances of the specified predetermined word is less than the predetermined number of times. The information processing program according to claim 1 or 2 , wherein a computer executes a process of outputting the one or a plurality of co -occurrence words according to the number of occurrences. The information processing program according to claim 1 or 2 , wherein a computer executes a process of outputting the one or a plurality of co -occurrence words according to an increasing tendency of the number of occurrences. It is characterized in that a computer is made to execute a process of outputting the different words according to the number of occurrences of a word different from the one or a plurality of co - occurrence words included in a post containing each of the one or a plurality of co-occurrence words. The information processing program according to any one of claims 1 to 4 . Identify the number of occurrences of a given word for a particular event in multiple posts over a given time period
When the number of appearances of the specified predetermined word is equal to or greater than the predetermined number of times, one or a plurality of words different from the predetermined word are extracted from the posts containing the predetermined word.
The ratio of the number of appearances of each of the one or more words in the predetermined period to the number of appearances of the one or more words in a period different from the predetermined period is the number of appearances of the predetermined word in a period different from the predetermined period. When the rate of the number of appearances of the predetermined word in the predetermined period is the same as or similar to that of the above, one or more words are output.
The specific event is estimated using the predetermined word and the output one or more words.
An information processing program characterized by having a computer execute processing. When a temporary increase in message transmission including a predetermined word related to a specific event is detected, the predetermined word and the predetermined word are co-occurred for the message transmitted in the period corresponding to the temporary increase. Extract co -occurrence words,
The predetermined word and the extracted co- occurrence word are output .
The specific event is estimated using the predetermined word and the output co-occurrence word.
A message analysis program characterized by having a computer execute processing. When a temporary increase in message transmission including a predetermined word related to a specific event is detected, the predetermined word and the predetermined word are co-occurred for the message transmitted in the period corresponding to the temporary increase. Extract the co -occurrence word, which is a co-occurrence word and is an adjective,
The period is evaluated based on the transmission status of the message inside and outside the period including the co-occurrence word which is the extracted adjective.
The specific event is estimated using the predetermined word and the co-occurrence word.
A message analysis program characterized by having a computer execute processing. A specific part that specifies the number of occurrences of a predetermined word for a specific event included in multiple posts in a predetermined period,
When the number of appearances of the specified predetermined word is equal to or greater than the predetermined number of times, the post containing the predetermined word is different from the predetermined word and co-occurs the predetermined word. An extractor that extracts multiple co -occurrence words,
Depending on whether or not the number of appearances of each of the extracted one or more co -occurrence words is greater than or equal to a specific threshold value as compared with the number of appearances in a period different from the predetermined period, the one or more co -occurrence words may be selected. The output section to output and
An estimation unit that estimates the specific event using the predetermined word and the output one or more co-occurrence words.
An information processing device characterized by having. The computer
Identify the number of occurrences of a given word for a particular event in multiple posts over a given time period
When the number of appearances of the specified predetermined word is equal to or greater than the predetermined number of times, one or a plurality of posts containing the predetermined word, which are different from the predetermined word and co-occurrence the predetermined word. Extract the co -occurrence word of
Depending on whether or not the number of appearances of each of the extracted one or more co -occurrence words is greater than or equal to a specific threshold value as compared with the number of appearances in a period different from the predetermined period, the one or more co -occurrence words may be selected. Output and
The specific event is estimated using the predetermined word and the output one or more co-occurrence words.
An information processing method characterized by executing processing.

Images