JP2006172424A - Information processor and information processing method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an information processor and information processing method capable of keeping a score on each topic of each channel and comparing between the channels wherein the topic, which is newer and occurs more heavily in recent days, scores a higher value, and processing at a high speed when a message is sequentially added. <P>SOLUTION: The information processor includes a message holding part for holding the message with a time related, and a score calculating part for dividing each of the messages held by the message holding part into words, extracting a string of the words matching to a pattern previously given from a string of the divided words from each of the messages as the topic, and obtaining the score on each of the topics with a specified time as a reference. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an information processing apparatus and an information processing method related to a communication system such as a bulletin board system and a chat system.

  In bulletin board systems and chat systems, it is required to create a place for communication that attracts users who want to access frequently. In order to attract users to the place of communication, useful information, new topics, and exchanges of interesting and entertaining messages need to occur on the spot.

  In order to generate the above-mentioned useful information in the field, it is important to attract many users who have the same interest in the channel through which messages of users having the same interest flow. This is because information sources can be diversified, and many new information is provided by people's daily discovery of information, proposals for various solutions, discussions and conversations from various viewpoints, etc. occur.

  As a conventional technique for attracting a user to a channel, a technique referred to Non-Patent Document 1 is known. For each channel, this technique selects a topic from the channel and presents it to the user as the topic of the channel.

  Specifically, a sequence of words that match the prepared part-of-speech pattern is regarded as a topic and extracted from the message in the channel. For example, the prepared pattern is a pattern that attracts the user's interest, such as a string of nouns or a character string in which nouns are connected by “no”. Then, a higher score is given to the extracted topic as it is generated more closely in the vicinity of the latest message of the channel, and a higher score is selected.

This method selects a topic that is new and is more closely generated near the latest message from the extracted topics in order, so that a topic that indicates a trend among users of the channel or a current user. It is useful for users, such as topics that are becoming more and more interesting, and can talk to multiple other users and present topics in channels that attract users' interest in terms of expression. Excellent for attracting users to each channel.
Megumi Ishii, Shuichi Nakawatase, Junji Tomita, "Proposal of Topic Extraction Method Using Noun Phrase and Word Momentum", Proc. Of the IPSJ 160th Natural Language Processing Study Group, 2004-NL-160, March 2004 5th, Information Lab Vol. 2004, no. 23, pp. 79-84

  In the above-described prior art, topic scores are obtained for each channel based on the latest message of the channel, and thus it is not possible to compare topics between channels. Therefore, it has the following problems to be solved.

  For example, the topic of a channel for which a new message has not occurred for a month and the topic of a channel for which new messages are continuously generated may have the same score. Therefore, channels with topics that are out of date and useless, such as how to obtain advance tickets for concerts held a month ago, and other users who have not sent messages to those channels recently. Led users to channels that could not talk to.

  Further, the topic of a channel with a short message generation interval and the topic of a channel with a long generation interval may have the same score. For example, a topic generated in a channel in which several tens of messages are generated in a few minutes and a topic generated in a channel in which only about one message is generated per day are represented by occurrence patterns (for example, the latest message has the same posting time 2 In one channel, if the same pattern appears for the first time only in the latest three messages, the same score is obtained. For this reason, messages are not easily generated, and users are guided to channels that take time to respond.

  The present invention has been made in view of the above-mentioned problems, and the object of the present invention is to compare between channels that have higher values as newer and more recent occurrences for each topic of each channel. An object of the present invention is to provide an information processing apparatus and an information processing method that can perform scoring and that perform such processing at high speed even when messages are sequentially added.

  In order to solve the problem, the present invention according to claim 1, the message holding unit holding a message associated with a time, and dividing each message held in the message holding unit into words, A score calculation unit that extracts a word sequence that matches a pattern given in advance from the divided word sequence as a topic from each message, and obtains a score based on a time specified for each topic; The score calculation unit has a difference between the designated time and the time associated with the message for each topic of each message as a denominator component, and the associated time is obtained from the message. The density of the topic is obtained using an expression having the number of new messages including the topic as a component of the numerator, and the representative value of the density is used as the topic score for each topic. Mel.

  In such a claim 1, the information processing apparatus includes a message holding unit and a score calculation unit, the message holding unit holds a message associated with a time, and the score calculation unit is provided in the message holding unit. Each retained message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from each message, and specified for each topic of each message. Using a formula that has the difference between the current time and the time associated with the message as the denominator component, and the associated time is newer than the message and the number of messages containing the topic as the numerator component. The density of the topic is obtained, and the representative value of the density is obtained as the topic score for each topic.

  The score calculation unit has a difference between the time specified for each topic of each message and the time associated with the message as a denominator component, and the associated time is newer than the message and the topic is selected. By calculating the density of the topic using an expression having the number of messages to be included in the numerator component, and obtaining the representative value of the density as the topic score for each topic, the score calculation reference point is made the same between channels. In addition, since the message interval can be measured on a common scale between channels, it is possible to obtain a score that can be compared between channels.

  According to a second aspect of the present invention, there is provided a message holding unit that holds a message associated with a time, and each message held in the message holding unit is divided into words. A sequence of words that match a predetermined pattern from the sequence is extracted as a topic from each message, and for each topic of each message, the time associated with the message and the associated time are extracted from the message. An index creating unit associating the number of new messages including the topic, an index holding unit holding the result of the index creating unit, and a time specified for the topic held in the index holding unit as a reference A score calculation unit for obtaining a score obtained by the score calculation unit, For each topic of each message, the difference between the specified time and the time associated with the message has a denominator component, and the associated time is newer than the message and includes the topic The density of the topic is obtained using an expression having the number of messages as a component of the numerator, and a representative value of the density is obtained as a topic score for each topic.

  In such a second aspect, the information processing apparatus includes a message holding unit, an index creating unit, an index holding unit, and a score calculating unit, wherein the message holding holds a message associated with a time, and the index creating unit Divides each message held in the message holding unit into words, extracts a word string that matches a predetermined pattern from the divided word strings as a topic from each message, For each topic, the time associated with the message and the number of messages whose associated time is newer than the message and includes the topic, the index holding unit holds the result of the index creating unit , The score calculation unit refers to the index holding unit, and each message For each topic, the difference between the specified time and the time associated with the message is a denominator component, and the number of messages with the associated time that is newer than the message and includes the topic The density of the topic is obtained using an expression of the component, and the representative value of the density is obtained as the topic score for each topic.

  The index creation unit obtains the values used for calculating the topic and the topic score, and the index holding unit stores the values obtained by the index creation unit, so that the process of obtaining these values each time the time is specified becomes unnecessary. . Further, the score calculation unit refers to the index holding, and uses the difference between the time specified for each topic of each message and the time associated with the message as a denominator component, and the associated time By calculating the density of the topic using an equation having the number of messages that are newer than the message and including the topic as a component of the numerator, and obtaining a representative value of the density as a topic score for each topic, The same score calculation reference point and the message interval can be measured on a common scale between channels. As described above, a score that can be compared between channels can be obtained at high speed according to the designation of time.

  Further, the present invention according to claim 3 holds topic information for each topic of each message in a maximum of N messages in ascending order of associated time, and the topic information is associated with the topic and the message. An index holding unit characterized in that it includes the number of messages that are newer than the message and include the topic, and the message associated with the time is input. The message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, and the extracted topic is held in the index holding unit. The number of the topic information having a topic that matches the topic that is being increased by 1 or the input A topic information in which the number of messages associated with a message and the number of messages including the topic are newer than the message and the associated time is created, added to the index holding unit, When topic information for N messages is stored in the index storage unit before, an index update unit that deletes topic information generated from a message with the oldest associated time, and the index storage unit A score calculation unit that obtains a score based on a specified time for a topic that is held, the score calculation unit refers to the index holding unit, and for each topic of each message The difference between the specified time and the time associated with the message is related to the denominator component. That time obtains the density of the topics using the equation with the number of messages to the components of the molecules comprising the new and the topic from the message, it obtains a representative value of the density for each topic as a score of the topic.

  In such a third aspect, the information processing apparatus is provided with an index holding unit, an index updating unit, and a score calculation unit, and the index holding unit is configured to store each message in a maximum of N messages in ascending order of associated time. The topic associated with the message for each topic and the associated time is newer than the message and holds topic information indicating the number of messages including the topic, and the index update unit receives a message, The input message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, and the index holding unit for each extracted topic The number of messages of topic information having a topic that matches the topic held in And generating topic information in which the time associated with the input message, the associated time is newer than the message and the number of messages including the topic is zero, and the index holding unit And when topic information for N messages is held in the index holding unit before the adding process, the topic information generated from the message with the oldest associated time is deleted, and the score The calculation unit refers to the index holding unit, and uses the difference between the designated time and the time associated with the message for each topic of each message as a denominator component, and the associated time is The topic using an expression that has the number of messages that are newer than the message and include the topic as a component of the numerator Calculated density, obtaining the representative value of the density as a score of the topic for each topic.

  The index holding unit holds values used for calculating topics and topic scores, and when the index update unit receives a message, the input message is divided into words, and the divided message is divided into words. Then, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, and the extracted topic matches the topic held in the index holding unit. The number of messages of topic information having a topic to be increased by one, and the time associated with the input message, the number of messages associated with the topic that is newer than the message and includes the topic Topic information is created, added to the index holding unit, and the number of messages in the index holding unit is set to N before the adding process. If the topic information is stored, the topic information generated from the message with the oldest associated time is deleted, so that all the information necessary for calculating the topic and topic score when adding a message is obtained. First, information is obtained only in a range that is affected by the addition of a message, so that a topic or a value used for topic score calculation can be obtained at high speed.

  By holding the value used by the index holding unit for calculating the topic or topic score, the process for obtaining the value used for obtaining the topic or score each time is specified becomes unnecessary. The score calculation unit refers to the index holding, and uses the difference between the time specified for each topic of each message and the time associated with the message as a denominator component, and the associated time is Score between channels by calculating the density of the topic using an equation with the number of messages that are newer than the message and including the topic as a component of the numerator, and obtaining the representative value of the density as the topic score for each topic. The calculation reference point is the same, and the message interval can be measured on a common scale between channels. As described above, even when messages are sequentially added, it is possible to obtain a score that can be compared between channels at high speed according to the designation of time.

  According to a fourth aspect of the present invention, there is provided a message holding unit for holding a message associated with a time, and dividing each message held in the message holding unit into words. A score calculation unit that extracts a sequence of words that match a pattern given in advance from the sequence as a topic from each message, and obtains a score based on a time specified for each topic, the score calculation unit Is the difference between the specified time for each topic of each message and the time associated with the message as a denominator component, and the time associated with each word constituting the topic Finds the value of the topic using an expression having the number of messages that are newer than the message and include the word as a component of the numerator, and the representative value of the value for each topic is the topic score. And determined.

  In such a fourth aspect, instead of the score calculation unit in the first aspect, each message is divided into words, and a string of words that match a pattern given in advance from the divided word string is assigned to each message. As a topic, the difference between the time specified for each topic in each message and the time associated with the message is used as a denominator component, and is associated with each word constituting the topic. A score calculation unit that obtains the value of the topic using an expression having the number of messages that are newer than the message and includes the word as a component of the numerator, and obtains the representative value of the value as the topic score for each topic Is provided. The score calculation unit of the invention 4 obtains the topic score by using the number of messages including the word of each word constituting the topic in a message associated with a newer time than the message in which the topic appears. After a topic is filed, topic scoring is possible even for a message that is expressed as a partial character string of the topic, such as an abbreviation, and the topic does not appear in the exact same form in the message.

  Further, the present invention according to claim 5, a message holding unit that holds a message associated with time, and each message held in the message holding unit is divided into words, and the divided words A sequence of words that match a pattern given in advance from the sequence is extracted as a topic from each message, and for each topic of each message, the time associated with the message and each word constituting the topic The index creation unit for associating the number of messages whose associated time is newer than the message and includes the word, the index retention unit for retaining the result of the index creation unit, and the index retention unit A score calculation unit that obtains a score based on a time designated for a topic, and the score calculation Refers to the index holding unit, and uses the difference between the specified time and the time associated with the message for each topic of each message as a denominator component, for each word constituting the topic. The value of the topic is obtained by using an expression having a numerator component of the number of messages having a newer time than the message and including the word, and the representative value of the value is set as the topic score for each topic. Ask.

  In such a fifth aspect, instead of the index holding unit, the index creation unit, and the score calculation unit in the second aspect, each message is divided into words, and a pattern given in advance from the divided word sequence is obtained. A string of matching words is extracted as a topic from each message, and for each topic of each message, there is a time associated with the message and a time associated with each word constituting the topic. An index creation unit that associates the number of messages that are newer than the message and include the word, an index storage unit that holds the result of the index creation unit, and the index storage unit, and is held by the index storage unit Each topic of each message is associated with the specified time and the message. Using a formula that has the difference in time as a denominator component, and for each word constituting the topic, the associated time is newer than the message and the number of messages including the word is a numerator component A score calculation unit for obtaining the topic value and obtaining a representative value of the value as a topic score is provided for each topic. In the index creation unit, the index holding unit, and the score calculation unit according to the fifth aspect, the topic associated with each topic of each message is related to each word constituting the topic and the time associated with the message of the word is new and By processing the number of messages containing words and holding the result of the index creation unit in the index holding unit, after the topic appears, the topic is expressed as a substring of the topic as an abbreviation Among them, high-speed topic scoring is possible even for messages that do not appear in exactly the same form.

  The present invention according to claim 6 holds topic information for each topic of each message in a maximum of N messages in the order of newest associated time, and the topic information is associated with the topic and the message. And an index holding unit characterized in that the time associated with each word includes the number of messages that are newer than the message and include the word. The inputted message is divided into words, a word string that matches a predetermined pattern is extracted as a topic from the divided word string, and the inputted message For each word included in the word, the mem- ber of the word of the topic information having the word held in the index holding unit. The number of sages is increased by 1, and the time associated with the input message for each extracted topic and the time associated with each word constituting the topic are Topic information that is newer than the message and includes 0 messages including the word is created and added to the index holding unit. Before the adding process, topic information for N messages is held in the index holding unit. If the index update unit deletes the topic information generated from the message with the oldest associated time, and the time specified for the topic held in the index holding unit A score calculation unit for obtaining a score, wherein the score calculation unit refers to the index holding unit and applies to each topic of each message; The difference between the specified time and the time associated with the message is a denominator component, and for each word constituting the topic, the associated time is newer than the message and the word The value of the topic is obtained using an expression having the number of messages including “” as a numerator component, and the representative value of the value is obtained as the topic score for each topic.

  In such a sixth aspect, instead of the index holding unit, the index update unit, and the score calculation unit in claim 3, the topic information for each topic of each message in the maximum N messages in the order of the associated time. The topic information includes the time associated with the topic and the message, the words constituting the topic, and the number of messages in which the associated time of each word is newer than the message and includes the word. When the message is input, the index holding unit characterized in that the input message is divided into words, and a word string that matches a predetermined pattern from the divided word strings Extracted as a topic, and for each word included in the input message, the index holding unit The number of messages of the word of the topic information having the word held is increased by 1, and the time associated with the input message and the topic are set for each extracted topic. For each constituent word, topic information is created with the associated time being newer than the message and the number of messages including the word is zero, added to the index holding unit, and before the adding process When topic information for N messages is stored in the index storage unit, refer to the index update unit that deletes topic information generated from the message with the oldest associated time, and the index storage unit , The designated time for each topic of each message held in the index holding unit and the message The associated time difference is the denominator component, and for each word constituting the topic, the associated time is newer than the message and the number of messages containing the word is the numerator component. A score calculation unit is provided for obtaining the value of the topic using an equation and obtaining the representative value of the value as the topic score for each topic.

  In the index creation unit, the index holding unit, and the score calculation unit according to claim 6, the time associated with the message of the word is new to each word constituting the topic for each topic of each message, and the word is After the topic has appeared by processing the number including the message and the index update unit updates one part instead of all the information in the index holding unit when adding a message, the partial character of the topic like an abbreviation The topics are represented as columns, and the topics do not appear in the same form in the message, and high-speed topic scoring is possible even for messages that are added sequentially.

  Further, the present invention described in claim 7 has a topic appearance number holding unit that holds the number of messages in which each topic appears in messages to be processed in any one of claims 1 to 3, The score of each topic is obtained by weighting with an expression having the number of messages for the topic in the topic appearance holding unit as a denominator component.

  Moreover, the present invention according to claim 8 has a word appearance number holding unit for holding the number of messages in which each word appears in the message to be processed in any one of claims 4 to 6, The score of each topic, the time associated with each word constituting the topic is newer than the message and the number of messages including the word is the number of messages for the word in the word appearance holding unit and the denominator component Weighted with the following formula.

  In such Claims 7 and 8, in any one of Claims 1 to 3 and Claims 4 to 6, each topic or each of the messages processed in the topic appearance number holding unit or the word appearance number holding unit The number of messages in which each word constituting the topic appears is retained, the score of each topic is calculated by the score calculation unit, the number of the messages for the topic is an element of the denominator, and each word constituting the topic Many messages are obtained by weighting the number of messages that have a newer associated time than the message and include the word with an expression having the number of messages for the word in the word appearance holding unit as a denominator component. The score of a general topic to be spoken can be reduced, and a rare topic can be prioritized.

  According to a ninth aspect of the present invention, there is provided a step of holding a message associated with a time, and dividing each held message into words, which are given in advance from the divided word strings. Extracting a sequence of words that match the pattern from each message as a topic, and obtaining a score based on a time designated for each topic, and obtaining the score includes: For each topic, the difference between the specified time and the time associated with the message is a denominator component, and the number of messages with the associated time that is newer than the message and includes the topic The density of the topic is obtained using an expression of the component, and the representative value of the density is obtained as the topic score for each topic.

  According to a tenth aspect of the present invention, there is provided a step of holding a message associated with a time, and dividing each of the held messages into words, which are given in advance from the divided word sequence. A string of words matching the pattern is extracted as a topic from each message, and for each topic of each message, the time associated with the message and the associated time are newer than the message and include the topic Associating the number of messages, retaining the associated result, and determining a score based on a specified time for a topic in the retained associated result. Then, the step of obtaining the score refers to the stored associated result and refers to the topic for each topic of each message. The difference between the received time and the time associated with the message is the denominator component, and the associated time is newer than the message and the number of messages containing the topic is the numerator component. Thus, the density of the topic is obtained, and a representative value of the density is obtained as a topic score for each topic.

  Further, the present invention according to claim 11 has a step of storing topic information for each topic of each message in a maximum of N messages in ascending order of associated time, and the topic information includes the topic and the topic The time associated with the message and the number of messages that are newer than the message and include the topic are included, and when the message associated with the time is input, the input A message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, and for each of the extracted topics, the retained topic and The number of the topic information having a matching topic is increased by 1, or associated with the input message. Create topic information in which the number of messages whose associated time is newer than the message and includes the topic is 0, add the topic information to the retained topic information, and retain the index before the adding process If topic information for the number N of messages is held in the part, the step of deleting the topic information generated from the message with the oldest associated time and the time specified for the held topic In the step of obtaining the score, the stored topic information is referred to, and a time designated for each topic of each message is associated with the message. A message that has a difference in the current time as a denominator component and the associated time is newer than the message and includes the topic Calculated density of the topics using the equation with the number to the components of the molecule, finding a representative value of the density as a score of the topic for each topic.

  The invention according to claim 12 is the step of holding a message associated with a time, and dividing each of the held messages into words, which is given in advance from the divided word sequence Extracting a sequence of words that match the pattern from each message as a topic, and obtaining a score based on a time designated for each topic, and in each step of obtaining the score, each topic of each message For each word constituting the topic, the associated time is newer than the message and the difference between the designated time and the time associated with the message is a denominator component. The value of the topic is obtained using an expression having the number of messages including words as a component of the numerator, and the representative value of the value is obtained as the topic score for each topic.

  Further, the present invention according to claim 13 is a step of holding a message associated with a time, and dividing each of the held messages into words, given in advance from the divided word sequence A sequence of words that match the pattern is extracted as a topic from each message, and each topic of the message is associated with the time associated with the message and each word constituting the topic. Associating the number of messages that are newer than the message and containing the word, retaining the associated result, and based on the time specified for the topic in the retained result Obtaining a score, wherein in the step of obtaining the score, each story of each message is referred to with reference to the stored associated result. For each word constituting the topic, the associated time is newer than the message and the difference between the designated time and the time associated with the message is a denominator component. The value of the topic is obtained using an expression having the number of messages including a word as a component, and the representative value of the value is obtained as the topic score for each topic.

  Further, the present invention according to claim 14 includes a step of holding topic information for each topic of each message in a maximum of N messages in ascending order of associated time, wherein the topic information includes the topic and the topic The time associated with the message, the words constituting the topic, and the associated time of each word include the number of messages that are newer than the message and include the word. When the message is input, the input message is divided into words, and a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, and the input message is extracted. For each word included, increase the number of messages for the word in the topic information having the held word by one, A message associated with the input message for each extracted topic, and a message associated with each word constituting the topic, the associated time being newer than the message and including the word If the topic information for the number N of messages is held in the index holding unit before the adding process, topic information with 0 as the number of messages is created and added to the topic information. Deleting the topic information generated from the oldest message, and obtaining a score based on the time specified for the topic of the retained topic information, and obtaining the score The step refers to the topic information held and associates the message with the designated time for each topic of each message. Using a formula that has the difference in time as a denominator component, and for each word constituting the topic, the associated time is newer than the message and the number of messages including the word is a numerator component A value of the topic is obtained, and a representative value of the value is obtained as a topic score for each topic.

  Further, the present invention according to claim 15 has the step of maintaining the number of topic appearances that holds the number of messages in which each topic appears in the messages to be processed. Then, the score of each topic is obtained by weighting with an expression having the number of messages for the topic in the topic appearance holding unit as a constituent element of the denominator.

  Further, the present invention according to claim 16 has the step of holding the number of messages in which each word appears in the message to be processed in any one of claims 12 to 14, and the score of each topic , The time associated with each word constituting the topic is newer than the message, and the number of messages including the word is weighted with an expression having the number of messages for the word in the word appearance holding unit as a denominator component And ask.

  According to the present invention, it is possible to perform scoring that can be compared between channels that have higher values as newer and more recent occurrences for each topic of each channel, and further, sequential messages are added Even in such a case, it is possible to provide an information processing apparatus and an information processing method that perform these processes at high speed.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

<First Embodiment>
FIG. 1 shows a first embodiment of an information processing apparatus according to the present invention. The information processing apparatus 1 in FIG. 1 includes a message holding unit 11 and a score calculation unit 12.

  The message holding unit 11 holds a message associated with a time. An example of the message holding unit 11 is shown in FIG.

  FIG. 20 is a configuration diagram of the score calculation unit 12. The score calculation unit 12 stores a time acquisition unit 201 that acquires time from the outside, a time storage unit 202 that stores the acquired time, a message order addition unit 203 that assigns an order to messages, and a message that is given an order. A topic-assigned message storage unit 204, a word dividing device 205 that divides the message into words, a pattern storage unit 206 that stores a pattern representing a topic, and a topic from each message in the message group, A topic extraction unit 207 that calculates the number of references in a message that is later in order than the message of each topic, a topic counter storage unit 208 that stores the number of references, and a score calculation unit 209 that calculates a topic score.

  FIG. 3 is an example of a processing flow of the score calculation unit 12. The processing flow will be described below.

<Score calculation section processing flow>
S31 order assignment processing (message order assignment section)
The messages held in the message holding unit are sorted according to the associated time, and the messages are assigned in order from oldest time to message 1... Message N, and the message group to which the order is given is assigned to the already assigned message storage unit. Output to.

  In S32 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit.

In S33 initial setting process (topic extraction unit),
(1) A message group is read from the order-assigned message storage unit.
(2) N = the number of read messages, and process counter i = 1.

In S34 topic extraction process completion determination process (topic extraction unit), if i> N (process complete for all messages), the process proceeds to S37.
In S35 topic extraction processing (topic extraction unit),
(1) The message i is output to the word segmentation device, and the list L = [[s1, t1], [s2] with the pair [sk, tk] of the word information tk such as the word sk and the part of speech or the presence / absence of dictionary registration , t2], ...] (s1, s2 in order from the first word of message i).
(2) A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, and a character string t for storing word information character strings are prepared, and R = [], s = "", T ="".
* “+” For list adds the element of the previous list as the forward element of the subsequent list. (Example) [a, b] + [c] = [a, b, c], [a, b] + [[c]] = [a, b, [c]], [[a, b], [c, d]] + [e] = [[a, b], [c, d], e]
+ For a character string appends the following character string to the end of the previous character string.
(Example) "ab" + "c" = "abc"
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to S36 (1).
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [s], s = “”, t = “”.
(11) As j = j + 1, the process proceeds to (5).

In S36 topic counter update processing (topic extraction unit),
(1) Create a counter with a value 0 for each topic (element of R) in message i and a topic counter for message i with the time associated with message i, and output to topic counter storage To do.
(2) Read the value of each topic (R element) in message i in the topic counter for messages 1 to i-1 in the topic counter storage unit, and add 1 to each of the read values Output to the topic counter storage unit.
(3) Set i = i + 1 and go to S34.

In S37 score calculation processing (score calculation unit),
(1) A specified time is read from the time storage unit, a value of each topic of each message, and a time associated with the topic counter of each message are read from the topic counter storage unit. The score is obtained by an expression having the difference between the designated time and the time associated with the message as the denominator component.
(2) For each topic, a representative value is determined from the score values of the topic obtained in S37 (1), and each topic and its representative value are output to the topic score value storage unit.

  In S32, a time T serving as a reference for score calculation is designated from the outside. For example, “13:50” is specified. In S35 (1), for example, the message is divided into words by dividing it with a character string registered in the dictionary using a morphological analyzer or a dictionary.

  In S35 (2) to (11), a part-of-speech pattern that is defined in advance using the result of S35 (1) (for example, a regular expression representing a string of nouns or a string in which nouns are connected by “no”). Noun (no ??) + noun) or a sequence of character strings registered in the dictionary, or a string of words that match a pattern in which the characters registered in the dictionary are connected by "no" are extracted as topics.

  In S36 (1) of FIG. 3, the following counters are created.

_{[13:40, [[w b,} 0], [w d, 0], [w e, 0]]]
Topical w _b in the message m _i of time associated now "13:40", w _d, if there is w _e, is S36 (2) At time "13:38" in FIG. 3 is associated If there is a counter of topic w _a , w _b , w _o with _a value of 0 for message m _i−1 , the counter of topic w _b is incremented by 1 to the following state.

_{[13:38, [[w a,} 0], [w b, 1], [w o, 0]]]
The score calculator of S37 (1) in FIG. 3, with respect to the topic w _j of the message m _i, the counter C _mi, has a component of _wj, associated with the specified the time T and the message m _i The topic score score _{mi, wj} is obtained by using a value obtained by an expression having a difference of a certain time t _mi as a denominator component. For example, the following formula can be used.

score _{mi, wj} = C _{mi, wj} / (T-t _mi )
Since a plurality of scores are obtained for one topic, for example, the score of the oldest message may be used as the score of the topic, or the highest score may be adopted as the score of the topic (S37 (2)). ).

  When the score of the oldest message is adopted, since the density of messages in which the topic appears until the specified time after the topic appears, the higher the score, the more the topic appears. It will be spoken closely by the designated time, and is suitable for providing as a hot topic.

  In addition, when the highest score is selected, the density is calculated in the section where each topic appears most densely starting from the specified time, so the one with the highest score is near the specified time. It is suitable for providing as a hot topic.

  In order to prioritize a topic whose associated time is newer than the message and has a large number of messages including the topic, the score is the number of messages whose associated time is newer than the message and includes the topic. Since the value of the counter to represent is a component of the denominator of the numerator and the difference between the time specified to reduce the freshness of the topic and the time associated with the message is also given to the component of the denominator, the message The expression is not limited to the above expression as long as the expression has a number as a numerator and the time difference as a denominator.

  For example, the counter of each topic of each message at time 310 in FIG.

m ₁ : [13:38, [[w _a , 1], [w _b , 3], [w _c , 1]]]
_{m 2: [13:40, [[} w b, 2], [w d, 0], [w e, 0]]]
m ₃ : [13:48, [[w _b , 1], [w _c , 0], [w _f , 1]]]
m ₄ : [13:49, [[w _a , 0], [w _b , 0], [w _f , 0]]]
In this case, the score of each topic of each message when time T = 13: 50 is designated is as follows.

  When the maximum score is selected for each topic, the topic score is as follows.

_{w a = 1/12, w} b = 1/2, w c = 1/12, w d = 0, w e = 0, w f = 1/2
In S37 (2) of FIG. 3, the topic and its score are output.

Here, when the messages m _{1 to} m ₄ have the same contents and the messages 12:38, 12:40, 12:48, and 12:49 are associated with each other, the message holding unit 11 holds the time. The topic scores for the 13:50 designation are:

_{w a = 1/72, w} b = 3/72, w c = 1/72, w d = 0, w e = 0, w f = 1/62
In addition, when the messages m _{1 to} m ₄ have the same contents and the messages 13:00, 13:10, 13:30, and 13:49 are associated with the messages, respectively, The topic scores for the 13:50 designation are:

_{w a = 1/50, w} b = 3/50, w c = 1/50, w d = 0, w e = 0, w f = 1/20
As described above, the information processing apparatus according to the present invention can give a higher score to a topic when a message is generated in a short period of time near a designated time even if the messages have the same contents. By providing the information processing apparatus of the present invention for each channel and calculating the topic score, it is possible to give a score that can be compared between channels to the topic.

  When there are a plurality of messages associated with the same time, for example, the order of the messages is determined using an input order or an identifier indicating the order of the messages, and the unit time is associated with the message. A score may be calculated by regarding a value obtained by sequentially adding a value obtained by equally dividing the number of messages in the associated time in order of messages as a time associated with each message. For example, the score can be calculated by replacing the time associated with the message with the sequentially added value in the message order assigning unit and performing the processes of S32 to S37.

<Second Embodiment>
FIG. 4 shows a second embodiment of the information processing apparatus according to the present invention. The message holding unit is the same as the message holding unit of the first embodiment. FIG. 21 is a configuration diagram of the index creation unit 41. FIG. 5 is an example of the processing flow of the index creation unit. The same symbols in the drawings represent the same devices and processes. In S51 of FIG. 5 (211 of FIG. 21), the following counter information of each topic of each message stored in the topic counter storage unit is output to the index holding unit.

[13:38, [[w _a , 1], [w _b , 3], [w _c , 1]]]
_{[13:40, [[w b,} 2], [w d, 0], [w e, 0]]]
[13:48, [[w _b , 1], [w _c , 0], [w _f , 1]]]
_{[13:49, [[w a,} 0], [w b, 0], [w f, 0]]]
FIG. 22 is a configuration example of the score calculation unit 43. FIG. 6 is an example of a processing flow of the score calculation unit 43. The processing flow will be described below.

<Score calculation section processing flow>
In the S61 stop check process, the control unit checks whether the process stop flag is on. If the flag is on, the control unit ends the process.

  In S62 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit. After outputting the time, the control unit is notified of the completion of output. The control unit instructs the score calculation unit to start processing. Thereafter, the process proceeds to S63.

In S63 score calculation processing (score calculation unit),
(1) Read the specified time from the time storage unit, the value of each topic of each message from the index holding unit, and the time associated with the topic counter of each message, respectively, and for each topic of each message, A score is obtained by an expression having a value as a component and having a difference between a designated time and a time associated with the message as a component of the denominator.

(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S63, and each topic and its representative value are output to the output storage unit. Thereafter, the control unit is notified of the completion of processing. Thereafter, the process proceeds to S61.

  By referring to the counter information held in the index holding unit 42 at the time of score calculation, it is not necessary to repeat the operation flows S31 and S33 to S36 of the score calculation unit of the first embodiment every time the time is specified. Thus, it is possible to quickly obtain a comparable score between channels of each topic at a specified time.

<Third Embodiment>
FIG. 7 shows a third embodiment of the information processing apparatus according to the present invention. FIG. 8 shows an example in which the index holding unit 72 holds topic information for up to three messages. The score calculation unit 73 is equivalent to the score calculation unit 43 of FIG. 4 of the second embodiment. FIG. 23 shows a configuration example of the index update unit. FIG. 9 shows an example of the processing flow of the index update unit 71.

The processing flow will be described below.

<Processing flow for updating the index>
In the S91 stop check process, the control unit checks whether the process stop flag is on, and if the flag is on, the process ends.

  In S92 message acquisition processing (message acquisition unit), a message is read from the input storage unit and output to the message storage unit. The control unit is notified of the completion of processing. The control unit instructs the topic extraction unit to start processing. Thereafter, the process proceeds to S93.

In S93 topic extraction processing (topic extraction unit),
(1) A message is read from the message storage unit and set as a message m.
(2) The message m is output to the word segmenting device, and the list L = [[s1, t1], [s2] having the pair [sk, tk] of the word information tk such as word sk and part of speech and dictionary registration , t2], ...] (s1, s2 in order from the first word of message i). A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, and a character string t for storing word information character strings are prepared, and R = [], s = "", T ="".
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to S94 (1).
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [s], S = “”, t = “”.
(11) As j = j + 1, the process proceeds to (5).

In S94 topic counter update processing (topic extraction unit),
(1) When a topic counter for less than N messages is held in the index holding unit, the process proceeds to (2). Otherwise, the process proceeds to (3).
(2) Delete the topic counter generated from the message with the oldest associated time. Thereafter, the process proceeds to (3).
(3) Read the value of each topic (element of R) in the message m in the topic counter of each message in the index holding unit, and output a value obtained by adding 1 to each of the read values to the topic counter storage unit .
(4) Create a counter with a value of 0 for each topic (element of R) in message m and a topic counter for message m with the time associated with message m, and add them to the index holding unit . Thereafter, the process proceeds to S91. N in S94 (1) is the upper limit of the number of messages held by the index holding unit.

  The index holding unit 72 holds topic counters for a maximum of three messages (N = 3 in S94 (1)), and when the index holding unit is in the state of FIG. 8, the topics wa, wb, The operation of the index update unit 71 when a message M including wf is input will be described below.

  In S94 (1), description will be made assuming that N = 3. The message acquisition unit reads the message M from the input storage unit, outputs the message M to the message storage unit, and notifies the control unit of the completion of processing. The control unit notifies the topic extraction unit of the start of processing (S91).

  The topic extraction unit reads a message from the message storage unit, divides the message into words (S93 (1), (2)), and extracts a topic (S93 (2) to (11)).

  Since the topic holding counter for the three messages is held in the index holding unit, the process of S94 (2) is performed. [13:38, [[wa, 0], [wb, 2], [wc, 1]]] with the oldest associated time is deleted. The counter of the topic counter held in the index holding unit for the topics wa, wb, wf included in the message m is incremented by 1 ((3) in S94). An example of information held in the index holding unit is shown below.

_{[13:40, [[w b,} 2], [w d, 0], [w e, 0]]]
[13:48, [[w _b , 1], [w _c , 0], [w _f , 1]]]
The topic information [13:49, [[w _a , 0], [w _b , 0], [w _f , 0]]] for the message M is added to the index holding unit in S94 (4) of FIG. The holding unit 72 is as follows.

_{[13:40, [[w b,} 2], [w d, 0], [w e, 0]]]
[13:48, [[w _b , 1], [w _c , 0], [w _f , 1]]]
_{[13:49, [[w a,} 0], [w b, 0], [w f, 0]]]
The score calculation unit 73 is the score calculation unit 43 of FIG. 4 according to the second embodiment. When a time is specified, the score calculation unit 73 refers to the index holding unit 72 and obtains and outputs a score for each topic.

  Each time the index update unit 71 adds a message, only the added message is word-divided and topic extraction is performed. Therefore, the messages are input to the information processing apparatus of the present invention for each channel in ascending order of the associated time. Thus, even when messages are sequentially added, it is possible to obtain a score that can be compared between channels at high speed according to the designation of time.

  If the index holding unit 72 is provided with a hash having a topic as a key and a list of pointers to counters of each topic of each message as a value, the index holding unit 72 for each topic included in the added message. It is possible to update the counter of the topic information held in the message at high speed, and to update the index when adding a message at a higher speed.

<Fourth embodiment>
FIG. 10 shows a fourth embodiment of the information processing apparatus according to the present invention. The score calculation unit 101 and the message holding unit 11 of the first embodiment are configured. FIG. 24 is a configuration example of the score calculation unit 101. The score calculation unit includes a time acquisition unit that acquires time from the outside, a message order assignment unit that assigns an order to messages, an order-added message storage unit that stores messages with an order, and a word that divides the message into words The topic is extracted from each message in the message group, and the dividing device, the pattern storage unit that stores the pattern representing the topic, and the number of references in the message in the order after the message of each word of each topic of each message A topic extraction unit with constituent word information to be obtained, a topic counter storage unit with constituent word information for storing the reference number, and a score calculation unit with constituent word information for calculating a topic score. FIG. 11 is an example of a processing flow of the score calculation unit 101. Details of the processing flow are shown below.

<Score calculation section processing flow>
In S111 order assignment processing (message order assignment section), the messages held in the message holding section are sorted according to the associated time, and the messages are ordered in the order of oldest as message 1 ... message N. The assigned message group is output to the order-assigned message storage unit.

  In S112 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit.

In S113 initial setting processing (topic extraction unit with constituent word information),
(1) A message group is read from the order-assigned message storage unit.
(2) N = the number of read messages, and process counter i = 1.
In S114 topic extraction processing completion determination processing (topic extraction unit with constituent word information), if i> N (processing is completed for all messages), the processing proceeds to S117.

In S115 topic extraction processing with constituent word information (topic extracting unit with constituent word information),
(1) The message i is output to the word segmentation device, and the list L = [[s1, t1], [s2] having a pair [sk, tk] of the word information tj such as the word sk and the part of speech or the presence / absence of dictionary registration , t2], ...] (s1, s2 in order from the first word of message i).
(2) A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, a character string t for storing word information character strings, and a word constituting each topic A list U and a list Q for storing words in the message are prepared, and R = [], s = "", t = "", U = [], Q = [].
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to S115.
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, U = U + [sj], and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [[s, U]], s = “”, t = “”, U = [].
(11) If sj εQ, the process proceeds to (13).
(12) Q = Q + [sj].
(13) As j = j + 1, the process proceeds to (5).

S116 In topic counter update processing with constituent word information (topic extracting unit with constituent word information),
(1) In each constituent word (R = [[s1, [u11, u12 ...]], [s2, [u21, ...]], ...] in message i) On the other hand, a topic counter with constituent word information is created for a message i having a counter with a value of 0 and the time associated with message i and the word included in message i, and output to the topic counter storage section with constituent word information To do.
(2) Read the value of each word (element of Q) in message i in the topic counter with constituent word information for message 1 to message i-1 in the topic counter storage unit with constituent word information, and each of the read values 1 is added to the topic counter storage unit with constituent word information.
(3) i = i + 1 and the process proceeds to S114.

S117 In the score calculation process with constituent word information (score calculator with constituent word information),
(1) The designated time from the time storage unit, the value of each word constituting each topic of each message from the topic counter storage unit with constituent word information, and the time associated with the topic counter with constituent word information for each message, respectively For each topic of each message, the counter value for the word constituting the topic is used as the numerator component, and the difference between the specified time and the time associated with the message is used as the denominator component. Find the score with the formula.
(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S117, and each topic and its representative value are output to the topic score value storage unit.
In S116 (1), the following counters and lists are created.
[13:40, [[w _b , [[b ₁₁ , 0], [b ₁₂ , 0], [b ₁₃ , 0]]], [w _d , [[d ₁₁ , 0], [d ₁₂ , 0]]],
_{[w e, [[e 11} , 0], [e 12, 0], [e 13, 0]]]],
_{[b 11, b 12, b} 13, d 12, d 12, e 11, e 12, e 13, o 21, o 22, o 11]]

Topical w _b in the message m _i of time associated now "13:40", has w _d, w _e, words constituting each _{topic, b 11, b 12, b} 13, d 11, d _12, e _11, a e _12, e _13, a message m words contained in the _i is _{b 11, b 12, b 13} , d 11, d 12, e 11, e 12, e 13, o 11, o _In the case of ₂₁ and o ₂₂ , in S116 (2), if there are counters of topics w _a , w _b , and w _{c having a} value of 0 for the message m _i−1 associated with the time “13:38”, the message since the word to _{m i b 11, b 12,} b 13 are included, the counter of the message m topic for _i-1 w _b word b _11, b _12, b ₁₃ increased by one to the following conditions.

_{[13:38, [[w a,} [[a 11, 0], [a 12, 0]]], [w b, [[b 11, 1], [b 12, 1], [b 13, 1]]],
[w _c , [[c ₁₁ , 0], [c ₁₂ , 0]]]],
_{[a 11, a 12, b} 11, b 12, b 13, c 11, c 12, o 11, o 12, o 13]]
Here, it is assumed that the time T is designated in S112 of FIG. The score calculation of S115 (1) in FIG. 11, has relative topic w _k messages m _i, the value C _mi of counters for word t _k ∈w _j constituting the _{topic, wj,} the _tk to the components of the molecule , using the values given by equation with the difference between the time t _mi associated with the specified the time T and the message m _i to the components of the denominator, score score _mi of the _topics, determine the _wj.

For example, the following formula can be used. | W _j | represents the number of words constituting w _j . The words configured here may simply be all the words obtained in the message division S115 (1), or only those registered in the dictionary. When morphological analysis is used, only nouns may be used, such as only nouns.

  As in the first embodiment, since a plurality of scores are obtained for one topic, for example, the topic score of the message with the oldest associated time is used as the topic score, or the highest score is obtained. You may employ | adopt as the score of the topic.

  For example, various counters for each topic of each message at the time of S115 (1) in FIG.

_{m 1: [13:38, [[} w a, [[a 11, 1], [a 12, 1]]],
[w _b , [b ₁₁ , 3], [b ₁₂ , 3], [b ₁₃ , 3]]],
[w _c , [[c ₁₁ , 1], [c ₁₂ , 1]]]],
_{[a 11, a 12, b} 11, b 12, b 13, c 11, c 12, o 11, o 12, o 13]]
m ₂ : [13:40, [[w _b , [[b ₁₁ , 2], [b ₁₂ , 2], [b ₁₃ , 2]]],
[w _d , [[d ₁₁ , 1], [d ₁₂ , 0]]],
_{[w e, [[e 11} , 0], [e 12, 0], [e 13, 0]]]],
_{[b 11, b 12, b} 13, d 11, d 12, e 11, e 12, e 13, o 21, o 22, o 11]]
m ₃ : [13:48, [[w _b , [[b ₁₁ , 1], [b ₁₂ , 1], [b ₁₃ , 1]]],
[w _c , [[c ₁₁ , c ₁₂ ]], [w _f , [f ₁₁ , f ₁₂ ]]],
_{[b 11, b 12, b} 13, c 11, c 12, f 11, f 12, d 11, o 31, o 22, o 13]]
_{m 4: [13:49, [[} w a, [[a 11, 0], [a 12, 0]]],
_{[w b, [[b 11} , b 12, b 13]], [w f, [f 11, f 12]]],
_{[a 11, a 12, b} 11, b 13, b 13, f 11, f 12, o 41, o 42, o 31]]
In this case, the score of each topic of each message when time T = 13: 50 is designated is as follows.

  When the maximum score is selected for each topic, the topic score is as follows.

_{w a = 1/12, w} b = 3/12, w c = 1/12, w d = 1/20, w e = 0, w f = 1/2
In S115 (2) of FIG. 11, the topic and its score are output.

Here, when the messages m _{1 to} m ₄ have the same contents and the messages 12:38, 12:40, 12:48, and 12:49 are associated with each other, the message holding unit 11 holds the time. The topic scores for the 13:50 designation are:

_{w a = 1/72, w} b = 3/72, w c = 1/72, w d = 1/140, w e = 0, w f = 1/62
In addition, when the messages m _{1 to} m ₄ have the same contents and the messages 13:00, 13:10, 13:30, and 13:49 are associated with the messages, respectively, The topic scores for the 13:50 designation are:

_{w a = 1/50, w} b = 3/50, w c = 1/50, w d = 1/80, w e = 0, w f = 1/20
As described above, the information processing apparatus according to the present invention can give a higher score to a topic when a message is generated in a short period of time near a designated time even if the messages have the same contents. By providing the information processing apparatus of the present invention for each channel and calculating the topic score, it is possible to give a score that can be compared between channels to the topic.

Furthermore, the message m _3, because it contains a word d ₁₁ constituting the topic w _d message m _2, in the first embodiment becomes possible scoring the topic w _d score was 0 ing. By using words, more detailed scoring is possible, and after a topic appears, a message that is expressed as a substring of the topic, such as an abbreviation, and the topic does not appear in the exact same form in the message Even topical scores are possible.

  In the above example, the topic score is obtained based on the average of the word counter values, but it may be only added without taking the average. If it is desired to increase the score for a topic composed of many words, it is preferable to add only.

  The word list of each message in the above embodiment is not necessary for the original processing, but is used for easy understanding of the operation example.

<Fifth embodiment>
FIG. 12 shows a fifth embodiment of an information processing apparatus according to the present invention. The message holding unit is the same as the message holding unit of the first embodiment. FIG. 25 is a configuration diagram of the index creation unit 121. FIG. 13 is an example of the processing flow of the index creation unit. The same symbols in the drawings represent the same devices and processes. In S131 of FIG. 13 (251 of FIG. 25), the following topic counter information of each word constituting each topic of each message stored in the topic counter storage unit is output to the index holding unit.

_{[13:38, [[w a,} [[a 11, 1], [a 12, 1]]], [w b, [b 11, 3], [b 12, 3], [b 13, 3 ]]],
[w _c , [[c ₁₁ , 1], [c ₁₂ , 1]]]],
_{[a 11, a 12, b} 11, b 12, b 13, c 11, c 12, o 11, o 12, o 13]]
[13:40, [[w _b , [[b ₁₁ , 2], [b ₁₂ , 2], [b ₁₃ , 2]]], [w _d , [[d ₁₁ , 1], [d ₁₂ , 0]]],
_{[w e, [[e 11} , 0], [e 12, 0], [e 13, 0]]]],
_{[b 11, b 12, b} 13, d 11, d 12, e 11, e 12, e 13, o 21, o 22, o 11]]
[13:48, [[w _b , [[b ₁₁ , 1], [b ₁₂ , 1], [b ₁₃ , 1]]],
[w _c , [[c ₁₁ , c ₁₂ ]], [w _f , [f ₁₁ , f ₁₂ ]]],
_{[b 11, b 12, b} 13, c 11, c 12, f 11, f 12, d 11, o 31, o 22, o 13]]
_{[13:49, [[w a,} [[a 11, 0], [a 12, 0]]],
_{[w b, [[b 11} , b 12, b 13]], [w f, [f 11, f 12]]],
_{[a 11, a 12, b} 11, b 13, b 13, f 11, f 12, o 41, o 42, o 31]]
FIG. 26 shows a configuration example of the score calculation unit 123. FIG. 14 is an example of the processing flow of the score calculation unit 123. The processing flow will be described below.

<Score calculation section processing flow>
In the S141 stop check process, the control unit checks whether the process stop flag is on. If the flag is on, the control unit ends the process.

  In S142 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit. After outputting the time, the control unit is notified of the completion of output. The control unit instructs the score calculation unit with constituent word information to start processing. Thereafter, the process proceeds to S143.

S143 In the score calculation process with constituent word information (score calculator with constituent word information),
(1) Read the specified time from the time storage unit, read the value of each word constituting each topic of each message from the index holding unit, and the time associated with the topic counter with constituent word information for each message, For each topic, the counter value for the word constituting the topic is a numerator component, and the score is expressed by an expression having the difference between the specified time and the time associated with the message as a denominator component. Ask.
(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S143, and each topic and its representative value are output to the output storage unit. Thereafter, the control unit is notified of the completion of processing. Thereafter, the process proceeds to S141.

  By referring to the counter information with constituent word information held in the index holding unit at the time of score calculation, each time the time is specified, the operation flows S111 and S113 of the score calculating unit with constituent word information of the fourth embodiment Since it is not necessary to repeat ~ S116, after a topic appears, it is specified even for a message that is expressed as a partial character string of the topic, such as an abbreviation, and the topic does not appear in the exact same form in the message. It is possible to quickly obtain a score that can be compared between channels of each topic for a given time.

<Sixth Embodiment>
FIG. 15 shows a sixth embodiment of the information processing apparatus according to the present invention. FIG. 16 shows an example when the index holding unit 152 holds topic counters with constituent word information for a maximum of three messages. The score calculation unit 153 is equivalent to the score calculation unit 123 of FIG. 12 according to the fifth embodiment. FIG. 17 shows an example of the processing flow of the index update unit 151. Details of the processing flow are shown below.

  FIG. 27 shows a configuration diagram of the index update unit 151. The index update unit 151 includes a control unit 271, a message acquisition unit 232, a message storage unit 233, a word division unit 205, a pattern storage unit 206, and a topic extraction unit 272 with constituent word information. . Further, an input storage unit 300 and an index holding unit 152 are connected to the index update unit 151.

<Processing flow for updating the index>
In the S171 stop check process, the control unit checks whether the process stop flag is on. If the flag is on, the control unit ends the process.

  In S172 message acquisition processing (message acquisition unit), a message is read from the input storage unit and output to the message storage unit. The control unit is notified of the completion of processing. The control unit instructs the topic extraction unit to start processing. Thereafter, the process proceeds to S173.

S173 In topic extraction processing with constituent word information (topic extracting section with constituent word information),
(1) A message is read from the message storage unit and set as a message m.
(2) The message m is output to the word segmenting device, and the list L = [[s1, t1], [s2] having a pair [sk, tk] of the word information tj such as the word sk and the part of speech or dictionary registration. , t2], ...] (s1, s2 in order from the first word of message i).
(3) A pattern is read from the pattern storage unit, and the pattern is set to P.
(4) A list R for storing topics in a message, a character string s for storing topic candidates, a character string t for storing word information character strings, and a word for composing each topic A list U and a list Q for storing words in the message are prepared, and R = [], s = "", t = "", U = [], Q = [].
(5) j = 1
(6) If j ≦ L is not satisfied, the process proceeds to S174.
(7) t = t + tj
(8) If there is a part that does not match P in t, the process proceeds to (10).
(9) s = s + sj, t = t + tj, U = U + [sj], and the process proceeds to (12).
(10) If s = “”, t = “” and the process proceeds to (12).
(11) R = R + [[s, U]], s = “”, t = “”, U = [].
(12) If sj ∈Q, the process proceeds to (13).
(13) Q = Q + [sj].
(14) As j = j + 1, the process proceeds to (6).

In S174 topic counter update processing with constituent word information (topic extracting unit with constituent word information),
(1) If the index holding unit holds topic counters with constituent word information for less than N messages, the process proceeds to (2). Otherwise, the process proceeds to (3).
(2) Delete the topic counter with constituent word information generated from the message with the oldest associated time. Thereafter, the process proceeds to (3).
(3) The value of each word (element of Q) in the message mi in the topic counter with constituent word information of each message in the index holding unit is read, and the value obtained by adding 1 to each of the read values is constituent word information Output to the topic counter storage unit.
(4) To each constituent word of each topic in message m (R = [[s1, [u11, u12 ...]], [s2, [u21, ...]], ...] uxy) On the other hand, a counter having a value of 0, a time counter associated with message i having a time associated with message i, and a word included in message m are created and added to the index holding unit. Thereafter, the process proceeds to S171.

  N in S174 (1) is the upper limit of the number of messages that the index holding unit holds the topic counter with constituent word information.

The index holding unit 152 holds topic information for a maximum of three messages. When the index holding unit 152 is in the state of FIG. 16, the topic w _a , w _b , w _f , the word a ₁₁ , associated with the time 13:49, _{a 12, b 11, b 12} , b 13, f 11, f 12, o 41, o 42, when receiving the message M containing o _31, illustrating the operation of the index update section 151 below.

The message M is divided into words (S173 (2) in FIG. 17), and topics are extracted (S173 (3) to (14) in FIG. 17). Since topic information for three messages, which is the upper limit, is held in the index holding unit 152, the process of S174 (2) is performed. Associated with that time is the oldest _{[13:38, [[w a,} [[a 11, 0], [a 12, 0]]], [w b, [b 11, 2], [b 12, 2], [b ₁₃ , 2]]], [w _c , [[c ₁₁ , 1], [c ₁₂ , 1]]]], [a ₁₁ , a ₁₂ , b ₁₁ , b ₁₂ , b ₁₃ , c ₁₁ , c ₁₂ , o ₁₁ , o ₁₂ , o ₁₃ ]] are deleted. The counter value of the word in each topic of each message held in the index holding unit 152 including the word included in the message M is incremented by 1 (S174 (3) in FIG. 17). An example of information held in the index holding unit 152 is shown below.

[13:40, [[w _b , [[b ₁₁ , 2], [b ₁₂ , 2], [b ₁₃ , 2]]], [w _d , [[d ₁₁ , 1], [d ₁₂ , 0]]],
_{[w e, [[e 11} , 0], [e 12, 0], [e 13, 0]]]],
_{[b 11, b 12, b} 13, d 11, d 12, e 11, e 12, e 13, o 21, o 22, o 11]]
[13:48, [[w _b , [[b ₁₁ , 1], [b ₁₂ , 1], [b ₁₃ , 1]]],
[w _c , [[c ₁₁ , c ₁₂ ]], [w _f , [f ₁₁ , f ₁₂ ]]],
_{[b 11, b 12, b} 13, c 11, c 12, f 11, f 12, d 11, o 31, o 22, o 13]]
Topic information [13:49 for the message M in S174 (4) in FIG. _{17, [[w a, [} [a 11, 0], [a 12, 0]]], [w b, [[b 11, b ₁₂ , b ₁₃ ]], [w _f , [f ₁₁ , f ₁₂ ]]],
_{[a 11, a 12, b} 11, b 13, b 13, f 11, f 12, o 41, o 42, o 31]]
Is added to the index holding unit, and the index holding unit 152 is as follows.

[13:40, [[w _b , [[b ₁₁ , 2], [b ₁₂ , 2], [b ₁₃ , 2]]], [w _d , [[d ₁₁ , 1], [d ₁₂ , 0]]],
_{[w e, [[e 11} , 0], [e 12, 0], [e 13, 0]]]],
_{[b 11, b 12, b} 13, d 11, d 12, e 11, e 12, e 13, o 21, o 22, o 11]]
[13:48, [[w _b , [[b ₁₁ , 1], [b ₁₂ , 1], [b ₁₃ , 1]]],
[w _c , [[c ₁₁ , c ₁₂ ]], [w _f , [f ₁₁ , f ₁₂ ]]],
_{[b 11, b 12, b} 13, c 11, c 12, f 11, f 12, d 11, o 31, o 22, o 13]]
_{[13:49, [[w a,} [[a 11, 0], [a 12, 0]]],
_{[w b, [[b 11} , b 12, b 13]], [w f, [f 11, f 12]]],
_{[a 11, a 12, b} 11, b 13, b 13, f 11, f 12, o 41, o 42, o 31]]
The score calculation unit 153 is the score calculation unit 123 of FIG. 12 according to the fifth embodiment. When a time is specified, the index holding unit 152 is referred to, and the score of each topic is obtained and output.

  Each time the index update unit 151 adds a message, only the added message is divided into words and the topic is extracted. Therefore, the messages are input for each channel to the information processing apparatus of the present invention in ascending order of the associated time. Thus, even when messages are added sequentially, the topic appears as a substring of the topic, such as an abbreviation, and the time of the message is not displayed in the exact same form in the message. It becomes possible to obtain a score that can be compared between channels at high speed according to the designation.

  In addition, if the index holding unit 152 is provided with a hash having a word as a key and a list of pointers to each word counter of each topic of each message as a value, an index is assigned to each word included in the added message. The topic information counter held in the holding unit 152 can be updated at high speed, and the index can be updated more quickly when a message is added.

<Seventh embodiment>
18A, 18B and 18C show a seventh embodiment of the information processing apparatus according to the present invention. When processing each message, the score calculation unit 182, the index creation unit 183, and the index update unit 186 count the number of messages (DF) in which each topic appears in addition to the processing of the first to third embodiments. Then, it is recorded in the topic appearance holding unit 181. 28 shows an example of the configuration of the score calculation unit 182, FIG. 29 shows the configuration of the index creation unit 183, FIG. 30 shows the score calculation unit 185, and FIG.

  An example of the processing flow of the score calculation unit 182, the index creation unit 183, the score calculation unit 185, and the index update unit 186 is shown below.

<< Score Calculation Unit 182 Process Flow >>
S182-1 (The first processing step in the score calculation unit 182 is denoted as S182-1. The same applies hereinafter.) In the ordering process (message ordering unit), the messages held in the message holding unit are related to each other. The messages are sorted in order from oldest time to message 1 ... message N, and the message group to which the order is assigned is output to the assigned message storage unit.

  In S182-2 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit.

In S182-3 initial setting processing (topic extraction unit),
(1) A message group is read from the order-assigned message storage unit.
(2) N = the number of read messages, and process counter i = 1.

  In S182-4 topic extraction processing completion determination processing (topic extraction unit), if i> N (processing is completed for all messages), the processing proceeds to S182-7.

In S182-5 topic extraction processing (topic extraction unit),
(1) The message i is output to the word segmentation device, and the list L = [[s1, t1], [s2] with the pair [sk, tk] of the word information tk such as the word sk and the part of speech or the presence / absence of dictionary registration , t2], ...] (s1, s2 in order from the first word of message i).
(2) A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, and a character string t for storing word information character strings are prepared, and R = [], s = "", T ="", Q = []. And
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to S36 (1).
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [s], s = “”, t = “”.
(11) As j = j + 1, the process proceeds to (5).
(12) Read the value for each element of R stored in the topic appearance holding unit 181, add 1 to each value, and set the value of each element of R recorded in the topic appearance holding unit 181 after the addition Update to each value. For the R element not recorded in the topic appearance holding unit 181, an area for storing the value of the element is created in the topic appearance holding unit 181, and a value 1 is output to the area.

In S182-6 topic counter update processing (topic extraction unit),
(1) Create a counter with a value 0 for each topic (element of R) in message i and a topic counter for message i with the time associated with message i, and output to topic counter storage To do.
(2) Read the value of each topic (R element) in message i in the topic counter for messages 1 to i-1 in the topic counter storage unit, and add 1 to each of the read values Output to the topic counter storage unit.
(3) Set i = i + 1, and the process proceeds to S182-4.

In S182-7 score calculation processing (score calculation unit),
(1) Read the designated time from the time storage unit, read the topic value of each message from the topic counter storage unit and the time associated with the topic counter for each message, and read the value of each topic from the topic appearance holding unit, For each topic of each message, the numerator is a value obtained by weighting the topic value read from the topic counter storage unit with an expression having the topic value read from the topic appearance holding unit as a denominator component. A score is obtained by an expression having a difference between a designated time and a time associated with the message as a component of the denominator.
(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S182-7, and each topic and its representative value are output to the topic score value storage unit.

<< Processing Flow of Index Creation Unit 183 >>
In the S183-1 order assignment process (message order assignment unit), the messages held in the message holding unit are sorted by the associated time, and the messages are sorted in order from oldest message to message 1 ... message N. The message group to which the order is assigned is output to the order-added message storage unit.

  In S183-2 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit.

In S182-3 initial setting processing (topic extraction unit),
(1) A message group is read from the order-assigned message storage unit.
(2) N = the number of read messages, and process counter i = 1.

  In S183-4 topic extraction processing completion determination processing (topic extraction unit), if i> N (processing is completed for all messages), the processing proceeds to S183-7.

In S183-5 topic extraction processing (topic extraction unit),
(1) The message i is output to the word segmentation device, and the list L = [[s1, t1], [s2] having a pair [sk, tk] of the word information tj such as the word sk and the part of speech or the presence / absence of dictionary registration , t2], ...] (s1, s2 in order from the first word of message i).
(2) A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, and a character string t for storing word information character strings are prepared, and R = [], s = "", T ="", Q = []. And
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to (12).
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [s], s = “”, t = “”.
(11) As j = j + 1, the process proceeds to (5).
(12) Read the value for each element of R stored in the topic appearance holding unit 181, add 1 to each value, and set the value of each element of R recorded in the topic appearance holding unit 181 after the addition Update to each value. For the R element not recorded in the topic appearance holding unit 181, an area for storing the value of the element is created in the topic appearance holding unit 181, and a value 1 is output to the area. Thereafter, the process proceeds to S183-6.

In S183-6 topic counter update processing (topic extraction unit),
(1) Create a counter with a value 0 for each topic (element of R) in message i and a topic counter for message i with the time associated with message i, and output to topic counter storage To do.
(2) Read the value of each topic (R element) in message i in the topic counter for messages 1 to i-1 in the topic counter storage unit, and add 1 to each of the read values Output to the topic counter storage unit.
(3) Set i = i + 1, and the process proceeds to S183-4.

  In S183-7 output processing (output unit), the value of each topic of each message and the time associated with each message are read from the topic counter storage unit and output to the index holding unit.

<< Score Calculation Unit 185 Process Flow >>
In the S185-1 stop check process, the control unit checks whether the process stop flag is on, and ends the process if the flag is on.

  In S185-2 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit. After outputting the time, the control unit is notified of the completion of output. The control unit instructs the score calculation unit to start processing. Thereafter, the process proceeds to S185-3.

In S185-3 score calculation processing (score calculation unit),
(1) Read the designated time from the time storage unit, read the value of each topic of each message from the index holding unit and the time associated with the topic counter for each message, and read the value of each topic from the topic appearance holding unit, For each topic of the message, the value of the topic read from the index holding unit is weighted with an expression having the topic value read from the topic appearance holding unit as a denominator component. The score is obtained by an expression having the difference between the specified time and the time associated with the message as an element in the denominator component.
(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S185-3, and each topic and its representative value are output to the output storage unit. Thereafter, the control unit is notified of the completion of processing. Thereafter, the process proceeds to S185-1.

<< Process Flow of Index Update Unit 186 >>
In the S186-1 stop check process, the control unit checks whether the process stop flag is on. If the flag is on, the control unit ends the process.

  In S186-2 message acquisition processing (message acquisition unit), a message is read from the input storage unit and output to the message storage unit. The control unit is notified of the completion of processing. The control unit instructs the topic extraction unit to start processing. Thereafter, the process proceeds to S186-3.

In S186-3 topic extraction processing (topic extraction unit),
(1) A message is read from the message storage unit and set as a message m.
(2) The message m is output to the word segmenting device, and the list L = [[s1, t1], [s2] having a pair [sk, tk] of the word information tj such as the word sk and the part of speech or dictionary registration. , t2], ...] (s1, s2 in order from the first word of message i).
A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, and a character string t for storing word information character strings are prepared, and R = [], s = "", T ="".
(4) j = 1
(5) If j ≤ number of elements, go to (12).
(6) t = t + tj
(7) If there is a part that does not match P in t, go to (9).
(8) Set s = s + sj, t = t + tj, and go to (11).
(9) If s = "", set t = "" and go to (11).
(10) R = R + [s], S = “”, t = “”.
(11) As j = j + 1, go to (5).
(12) Read the value for each element of R stored in the topic appearance holding unit 181, add 1 to each value, and set the value of each element of R recorded in the topic appearance holding unit 181 after the addition Update to each value. For the R element not recorded in the topic appearance holding unit 181, an area for storing the value of the element is created in the topic appearance holding unit 181, and a value 1 is output to the area. Thereafter, the process proceeds to S186-4.

In S186-4 topic counter update processing (topic extraction unit),
(1) When a topic counter for less than N messages is held in the index holding unit, the process proceeds to (2). Otherwise, the process proceeds to (3).
(2) Delete the topic counter generated from the message with the oldest associated time. Thereafter, the process proceeds to (3).
(3) Read the value of each topic (element of R) in the message m in the topic counter of each message in the index holding unit, and output a value obtained by adding 1 to each of the read values to the topic counter storage unit .
(4) Create a counter with a value of 0 for each topic (element of R) in message m and a topic counter for message m with the time associated with message m, and add them to the index holding unit . Thereafter, the process proceeds to S186-1.

  In the index update unit 186, when the topic counter is deleted for the old message after reaching the specified number of messages (S186-4 (2)), the value is read from the area for the topic included in the message to be deleted in the topic appearance holding unit, The respective values may be decremented by 1 and output to each area in the topic appearance holding unit.

  The score calculation units 182 and 185 obtain the value DFwj for the topic wj held in the topic appearance number holding unit 181 with respect to the topic wj of the message mi in 182-7 and 185-3. The score of each topic of each message with respect to the time T specified by weighting with a formula as a component is obtained. For example, the following formula can be used.

score _{mi, wj} = (C _{mi, wj} / DF _wj ) / (T-t _mi )
Similar to the first embodiment, when the highest score is adopted as the score of each topic, the topic score w _a = 1/12, w _b = 1/2, w _c = 1 in the first embodiment. _{/ 12, w d = 0,} w e = 0, w f = 1/2
Since the number of occurrences message is _{DF wa = 2, DF wb =} 4, DF wc = 2, DF wd = 1, DF we = 1, DF wf = 2,
_{w a = 1/24, w} b = 1/8, w c = 1/24, w d = 0, w e = 0, w f = 1/4
Thus, the score of the topic w _f that appears densely around the designated time and is a more unusual topic can be made higher than the score of the topic w _b that appears frequently.

  Note that a value given in advance may be used without updating the topic appearance holding unit 181.

<Eighth Embodiment>
19A, 19B, and 19C show an eighth embodiment of the information processing apparatus according to the present invention. When the score calculation unit 192, the index update unit 193, the index update unit 194, and the score calculation unit 195 process each message, the number of messages in which each word appears in addition to the processing of the fourth to sixth embodiments. (DF) is counted and recorded in the word appearance holding unit 191. 32 shows an example of the configuration of the score calculation unit 192, FIG. 33 shows an example of the configuration of the index creation unit 193, FIG. 34 shows an example of the configuration of the score calculation unit 195, and FIG.

  An example of the processing flow of the score calculation unit 192, the index creation unit 193, the score calculation unit 195, and the index update unit 196 is shown below.

<< Score Calculation Unit 192 Processing Flow >>
In S192-1 order assignment processing (message order assignment unit), the messages held in the message holding unit are sorted by the associated time, and the messages are ordered in the order of oldest message 1 ... message N. The message group to which the order is assigned is output to the order-added message storage unit.

  In S192-2 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit.

In S192-3 initial setting processing (topic extraction unit with constituent word information),
(1) A message group is read from the order-assigned message storage unit.
(2) N = the number of read messages, and process counter i = 1.

In S192-4 topic extraction processing completion determination processing (topic extraction unit with constituent word information),
If i> N (processing is completed for all messages), the process proceeds to S192-7.

In S192-5 topic extraction processing with constituent word information (topic extracting section with constituent word information),
(1) The message i is output to the word segmentation device, and the list L = [[s1, t1], [s2] having a pair [sk, tk] of the word information tj such as the word sk and the part of speech or the presence / absence of dictionary registration , t2], ...] (s1, s2 in order from the first word of message i).
(2) A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, a character string t for storing word information character strings, and a word constituting each topic A list U and a list Q for storing words in the message are prepared, and R = [], s = "", t = "", U = [], Q = [].
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to (14).
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, U = U + [sj], and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [[s, U]], s = “”, t = “”, U = [].
(11) If sj εQ, the process proceeds to (13).
(12) Q = Q + [sj].
(13) As j = j + 1, the process proceeds to (5).
(14) Read the value for each element of Q stored in the word appearance holding unit 191, add 1 to each value, and set the value of each element of Q recorded in the word appearance holding unit 191 to the value after the addition Update to each value. For an element of Q that is not recorded in the word appearance holding unit 191, an area for storing the value of the element is created in the word appearance holding unit 191, and a value 1 is output to the area. Thereafter, the process proceeds to S192-6.

In S192-6 topic counter update processing with constituent word information (topic extracting unit with constituent word information),
(1) In each constituent word (R = [[s1, [u11, u12 ...]], [s2, [u21, ...]], ...] in message i) On the other hand, a topic counter with constituent word information is created for a message i having a counter with a value of 0 and the time associated with message i and the word included in message i, and output to the topic counter storage section with constituent word information To do.
(2) Read the value of each word (element of Q) in message i in the topic counter with constituent word information for message 1 to message i-1 in the topic counter storage unit with constituent word information, and each of the read values 1 is added to the topic counter storage unit with constituent word information.
(3) Set i = i + 1, and the process proceeds to S192-4.

In the S192-7 score calculation process with constituent word information (score calculator with constituent word information),
(1) The designated time from the time storage unit, the value of each word constituting each topic of each message from the topic counter storage unit with constituent word information, and the time associated with the topic counter with constituent word information for each message, The value of each word is read from the word appearance holding unit, and the value of the word constituting the topic read from the topic counter storage unit with constituent word information is read from the word appearance holding unit for each topic of each message. A value obtained by weighting the value of a word constituting the topic with an expression having a denominator component as a denominator component is used as a numerator component, and the difference between the designated time and the time associated with the message is configured as a denominator. Find the score with the expression of the element.
(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S117, and each topic and its representative value are output to the topic score value storage unit.

<< Process Flow of Index Creation Unit 193 >>
In S193-1 order assigning process (message order assigning unit), the messages held in the message holding unit are sorted according to the associated time, and the messages are ordered in the order of oldest message 1 ... message N. The message group to which the order is assigned is output to the order-added message storage unit.

  In S193-2 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit.

In S193-3 initial setting processing (topic extraction unit with constituent word information),
(1) A message group is read from the order-assigned message storage unit.
(2) N = the number of read messages, and process counter i = 1.

  In S193-4 topic extraction processing completion determination processing (topic extraction unit with constituent word information), if i> N (processing is completed for all messages), the processing proceeds to S193-7.

In S193-5 topic extraction processing with constituent word information (topic extraction unit with constituent word information),
(1) Output message i to a word segmenter, list L = [[s1, t1], [s2] with word [sk], [part of speech], word information such as presence / absence of dictionary registration, etc. , t2], ...] (s1, s2 in order from the first word of message i).
(2) A pattern is read from the pattern storage unit, and the pattern is set to P.
(3) A list R for storing topics in a message, a character string s for storing topic candidates, a character string t for storing word information character strings, and a word constituting each topic A list U and a list Q for storing words in the message are prepared, and R = [], s = "", t = "", U = [], Q = [].
(4) j = 1
(5) If j ≦ L is not satisfied, the process proceeds to (14).
(6) t = t + tj
(7) If there is a part that does not match P in t, the process proceeds to (9).
(8) s = s + sj, t = t + tj, U = U + [sj], and the process proceeds to (11).
(9) If s = “”, the process proceeds to (11) with t = “”.
(10) R = R + [[s, U]], s = “”, t = “”, U = [].
(11) If sj εQ, the process proceeds to (13).
(12) Q = Q + [sj].
(13) As j = j + 1, the process proceeds to (5).
(14) Read the value for each element of Q stored in the word appearance holding unit 191, add 1 to each value, and set the value of each element of Q recorded in the word appearance holding unit 191 to the value after the addition Update to each value. For an element of Q that is not recorded in the word appearance holding unit 191, an area for storing the value of the element is created in the word appearance holding unit 191, and a value 1 is output to the area. Thereafter, the process proceeds to S193-6.

In S193-6 topic counter update processing with constituent word information (topic extracting unit with constituent word information),
(1) In each constituent word (R = [[s1, [u11, u12 ...]], [s2, [u21, ...]], ...] in message i) On the other hand, a topic counter with constituent word information is created for a message i having a counter with a value of 0 and the time associated with message i and the word included in message i, and output to the topic counter storage section with constituent word information To do.
(2) Read the value of each word (Q element) in message i in the topic counter with constituent word information for message i-1 in the topic counter storage unit with constituent word information, and for each of the read values The value added by 1 is output to the topic counter storage unit with constituent word information.
(3) Set i = i + 1, and the process proceeds to S193-4.

  In S193-7 output processing (output unit), each topic of each message, the value of its constituent word, and the time associated with each message are read from the topic counter storage unit with constituent word information and output to the index holding unit.

<< Score Calculation Unit 195 Process Flow >>
In the S195-1 stop check process, the control unit checks whether the process stop flag is on. If the flag is on, the control unit ends the process.

  In S195-2 designated time acquisition processing (time acquisition unit), the time specified from the outside is read from the input storage unit, and the time is output to the time storage unit. After outputting the time, the control unit is notified of the completion of output. The control unit instructs the score calculation unit with constituent word information to start processing. Thereafter, the process proceeds to S195-3.

In S195-3 score calculation processing with constituent word information (score calculating section with constituent word information),
(1) The designated time from the time storage unit, the value of each word constituting each topic of each message from the topic counter storage unit with constituent word information, and the time associated with the topic counter with constituent word information for each message, The value of each word is read from the word appearance holding unit, and the value of the word constituting the topic read from the topic counter storage unit with constituent word information is read from the word appearance holding unit for each topic of each message. A value obtained by weighting the value of a word constituting the topic with an expression having a denominator component as a denominator component is used as a numerator component, and the difference between the designated time and the time associated with the message is configured as a denominator. Find the score with the expression of the element.
(2) For each topic, a representative value is determined from the score values of the topic obtained in (1) of S195-3, and each topic and its representative value are output to the output storage unit. Thereafter, the control unit is notified of the completion of processing. Thereafter, the process proceeds to S195-1.

<< Process Flow of Index Update Unit 196 >>
In the S196-1 stop check process, the control unit checks whether the process stop flag is on, and ends the process if the flag is on.

  In S196-2 message acquisition processing (message acquisition unit), a message is read from the input storage unit and output to the message storage unit. The control unit is notified of the completion of processing. The control unit instructs the topic extraction unit to start processing. Thereafter, the process proceeds to S196-3.

In S196-3 topic extraction processing with constituent word information (topic extracting section with constituent word information),
(1) A message is read from the message storage unit and set as a message m.
(2) The message m is output to the word segmenting device, and the list L = [[s1, t1], [s2] having a pair [sk, tk] of the word information tj such as the word sk and the part of speech or dictionary registration. , t2], ...] (s1, s2 in order from the first word of message i).
(3) A pattern is read from the pattern storage unit, and the pattern is set to P.
(4) A list R for storing topics in a message, a character string s for storing topic candidates, a character string t for storing word information character strings, and a word for composing each topic A list U and a list Q for storing words in the message are prepared, and R = [], s = "", t = "", U = [], Q = [].
(5) j = 1
(6) If j ≦ L, the process proceeds to (15).
(7) t = t + tj
(8) If there is a part that does not match P in t, the process proceeds to (10).
(9) s = s + sj, t = t + tj, U = U + [sj], and the process proceeds to (12).
(10) If s = “”, t = “” and the process proceeds to (12).
(11) R = R + [[s, U]], s = “”, t = “”, U = [].
(12) If sj ∈Q, the process proceeds to (13).
(13) Q = Q + [sj].
(14) As j = j + 1, the process proceeds to (6).
(15) Read the value for each element of Q stored in the word appearance holding unit 191, add 1 to each value, and set the value of each element of Q recorded in the word appearance holding unit 191 to the value after the addition Update to each value. For an element of Q that is not recorded in the word appearance holding unit 191, an area for storing the value of the element is created in the word appearance holding unit 191, and a value 1 is output to the area. Thereafter, the process proceeds to S196-4.

S196-4 Topic counter update process with constituent word information (topic extraction unit with constituent word information)
(1) If the index holding unit holds topic counters with constituent word information for less than N messages, the process proceeds to (2). Otherwise, the process proceeds to (3).
(2) Delete the topic counter with constituent word information generated from the message with the oldest associated time. The process proceeds to (3).
(3) The value of each word (element of Q) in the message mi in the topic counter with constituent word information of each message in the index holding unit is read, and the value obtained by adding 1 to each of the read values is constituent word information Output to the topic counter storage unit.
(4) To each constituent word of each topic in message m (R = [[s1, [u11, u12 ...]], [s2, [u21, ...]], ...] uxy) On the other hand, a counter having a value of 0, a time counter associated with message i having a time associated with message i, and a word included in message m are created and added to the index holding unit. The process proceeds to S196-1.

  In the index updating unit 196, when topic information for an old message is deleted after reaching the specified number of messages, the number of words included in the message to be deleted may be decremented by one.

Score calculation unit 192,195, to the topic w _j of the message m _i in 192-7,195-3, the counter C _mi for words constituting the topic of each topic for each _{message, wj,} the _tk The score of each topic of each message for the specified time T is obtained by weighting the value DF _tk for the word t _k held in the word appearance holding unit 191 with an expression having a denominator component. For example, the following formula can be used.

  For example, a case where a message including the following topics and words is targeted will be described.

_{[13:40, [[w a,} [[a 11, 3], [a 12, 2]]], [w b, [[b 11, 1], [b 12, 0]]]], [ _{a 11, a 12, b 11} , b 12, o 21, o 22,]]
_{[13:45, [[w a,} [[a 11, 3], [a 12, 2]]]], [a 11, a 12, o 11, o 12,]]
_{[13:48, [a 11, b} 12, o 31, o 32,]]
_{[13:49, [[w a,} [[a 11, 0], [a 12, 0]]]], [a 11, a 12, b 11, o 11, o 12,]]
When “13:50” is designated as the time T, when the highest score is adopted as the score of each topic, the number of appearance messages of words of the following words constituting the topic DF _a11 = 4, DF _a13 = 3, DF _b11 = 2 and DF _b12 = 2
The topic score is w _a = 1/12, w _b = 1/10.
It becomes.

From the above, according to the present invention, it appears densely around the specified time, is a more unusual topic, and after the topic appears, the topic is represented in the message as a substring of the topic as an abbreviation. It is possible to make the score of the topic w _b that is a topic that does not appear in the same form higher than the topic w _a that frequently appears.

  Note that a value given in advance may be used without updating the word appearance holding unit 191.

  According to the embodiment of the present invention described above, the topic scores can be compared between channels, so a list of topics arranged in descending order of scores is presented to the user for all channel topics. Or presenting a list of channels arranged in the order of topics with high scores to the user, it is possible to give priority to channels that can talk to multiple users on useful and fresh topics. By visiting the presented channel, users can obtain useful information related to signs of epidemic and movements in the world. You can also spend a fulfilling time watching the heated conversation and participating in the conversation.

The structure for implement | achieving the 1st Embodiment of this invention is shown. The data hold | maintained at the message holding | maintenance part in the 1st Embodiment of this invention are shown. The processing flow of the score calculation part in the 1st Embodiment of this invention is shown. The structure for implement | achieving the 2nd Embodiment of this invention is shown. The processing flow of the index preparation part in the 2nd Embodiment of this invention is shown. The processing flow of the score calculation part in the 2nd Embodiment of this invention is shown. The structure for implement | achieving the 3rd Embodiment of this invention is shown. The data which the index holding part in the 3rd Embodiment of this invention hold | maintains are shown. The processing flow of the index update part in the 3rd Embodiment of this invention is shown. The structure for implement | achieving the 4th Embodiment of this invention is shown. The processing flow of the score calculation part in the 4th Embodiment of this invention is shown. The structure for implement | achieving the 5th Embodiment of this invention is shown. The processing flow of the index preparation part in the 5th Embodiment of this invention is shown. The processing flow of the score calculation part in the 5th Embodiment of this invention is shown. The structure for implement | achieving the 6th Embodiment of this invention is shown. The index holding unit in the sixth embodiment of the present invention indicates data to be held. The processing flow of the index update part in the 6th Embodiment of this invention is shown. The structure for implement | achieving the 7th Embodiment of this invention is shown. The structure for implement | achieving the 8th Embodiment of this invention is shown. The block diagram of the score calculation part which concerns on embodiment is shown. The block diagram of the index preparation part which concerns on embodiment is shown. The structural example of the score calculation part which concerns on embodiment is shown. The structural example of the index update part which concerns on embodiment is shown. The structural example of the score calculation part which concerns on embodiment is shown. The block diagram of the index preparation part which concerns on embodiment is shown. The structural example of the score calculation part which concerns on embodiment is shown. The block diagram of the index update part which concerns on embodiment is shown. The score calculation part which concerns on embodiment is shown. The index production part which concerns on embodiment is shown. The score calculation part which concerns on embodiment is shown. The structural example of the index update part which concerns on embodiment is shown. The score calculation part which concerns on embodiment is shown. The index production part which concerns on embodiment is shown. The score calculation part which concerns on embodiment is shown. An example of the structure of the index update part which concerns on embodiment is shown.

Explanation of symbols

1, 4, 7, 10, 12, 15 Information processing device 11 Message holding unit 12, 43, 73, 101, 123, 153, 182, 185, 192, 195 Score calculation unit 41, 72, 121, 183, 193 Index Creation unit 42, 72, 122, 152 Index holding unit 151, 186, 196 Index update unit 181 Topic appearance holding unit 191 Word appearance holding unit

Claims (16)

A message holding unit for holding a message associated with a time;
Each message held in the message holding unit is divided into words, a word sequence that matches a predetermined pattern is extracted from each message as a topic from the divided word sequence, and for each topic A score calculation unit that obtains a score based on the specified time,
The score calculation unit has a difference between the specified time and the time associated with the message for each topic of each message as a denominator component, and the associated time is newer than the message and An information processing apparatus characterized in that a density of a topic is obtained using an expression having the number of messages including the topic as a component of a numerator, and a representative value of the density is obtained as a topic score for each topic. A message holding unit for holding a message associated with a time;
Dividing each message held in the message holding unit into words, extracting a word sequence that matches a predetermined pattern from the divided word sequence as a topic from each message, each message of each message An index creating unit for associating a topic with a time associated with the message and a number of messages that are newer than the message and include the topic;
An index holding unit for holding the result of the index creation unit;
A score calculation unit that obtains a score based on the time specified for the topic held in the index holding unit,
The score calculation unit refers to the index holding unit, and is associated with each topic of each message by using a difference between the designated time and the time associated with the message as a denominator component. Obtaining the density of the topic using an expression having the number of messages that are newer than the message and including the topic as a component of the numerator, and obtaining a representative value of the density as a topic score for each topic. Information processing apparatus. The topic information for each topic of each message in a maximum of N messages in ascending order of the associated time is held, and the topic information is related to the topic, the time associated with the message, and the associated time. An index holding unit characterized by including the number of messages that are newer and include the topic;
When a message associated with a time is input, the input message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, For each extracted topic, the number of messages of topic information having a topic that matches the topic held in the index holding unit is increased by 1, or associated with the input message. Topic information in which the number of messages whose time and the associated time are newer than the message and include the topic is 0 is created, added to the index holding unit, and the index holding unit before the adding process If the topic information for N messages is stored in the message, it is generated from the message with the oldest associated time. And the index update section that you want to delete the topic information,
A score calculation unit that obtains a score based on the time specified for the topic held in the index holding unit,
The score calculation unit refers to the index holding unit, and uses the difference between the time specified for each topic of each message and the time associated with the message as a denominator component, and the associated time The density of the topic is obtained using an expression having the number of messages newer than the message and including the topic as a numerator component, and the representative value of the density is obtained as the topic score for each topic. Information processing device. A message holding unit for holding a message associated with a time;
Each message held in the message holding unit is divided into words, a word sequence that matches a predetermined pattern is extracted from each message as a topic from the divided word sequence, and for each topic A score calculation unit for obtaining a score based on the specified time,
The score calculation unit has a difference between the specified time for each topic of each message and the time associated with the message as a denominator component, and associates with each word constituting the topic The value of the topic is obtained using an expression having the number of messages that are newer than the message and including the word as a component of the numerator, and the representative value of the value is obtained as the topic score for each topic. An information processing apparatus characterized by the above. A message holding unit for holding a message associated with a time;
Dividing each message held in the message holding unit into words, extracting a word sequence that matches a predetermined pattern from the divided word sequence as a topic from each message, each message of each message An index creating unit that associates the number of messages that are related to the topic with the time associated with the message and each word constituting the topic, and the associated time is newer than the message and includes the word;
An index holding unit for holding the result of the index creation unit;
A score calculation unit that obtains a score based on the time specified for the topic held in the index holding unit,
The score calculation unit refers to the index holding unit, and uses the difference between the designated time and the time associated with the message for each topic of each message as a denominator component to configure the topic. The time value associated with each word is newer than the message and the value of the topic is obtained using an expression having the number of messages including the word as a component of the numerator, and the representative value of the value is determined for each topic. An information processing apparatus characterized by being obtained as a topic score. The topic information for each topic of each message in a maximum of N messages in ascending order of the associated time is held, and the topic information includes the time associated with the topic and the message, the words constituting the topic, and the topic An index holding unit characterized in that the associated time of each word is newer than the message and includes the number of messages including the word;
When a message associated with a time is input, the input message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, For each word included in the input message, the number of messages of the word in the topic information having the word held in the index holding unit is increased by 1, and each extracted topic On the other hand, for the time associated with the input message and each word constituting the topic, the associated time is newer than the message and the number of messages including the word is 0. Topic information is created, added to the index holding unit, and topic information for N messages is held in the index holding unit before the adding process. If it is, the index update section that you want to delete the topic information generated from the time associated with the oldest message,
A score calculation unit for obtaining a score based on a time specified for a topic held in the index holding unit;
The score calculation unit refers to the index holding unit, and uses the difference between the designated time and the time associated with the message for each topic of each message as a denominator component to configure the topic. For each word, the value of the topic is determined using an expression having the associated number of messages that is newer than the message and the number of messages including the word as a component of the numerator, and the representative value of the value for each topic An information processing apparatus characterized by obtaining a score as a topic score.   It has a topic appearance number holding unit that holds the number of messages in which each topic appears in the message to be processed, and the score of each topic is the number of messages for the topic of the topic appearance holding unit as a denominator component The information processing apparatus according to claim 1, wherein the information processing apparatus is obtained by weighting with an expression.   It has a word appearance number holding unit that holds the number of messages in which each word appears in the message to be processed, and the time associated with each word constituting the topic is newer than the message. The number of messages including the word is obtained by weighting the number of messages corresponding to the word in the word appearance holding unit with an expression having a denominator as a constituent element. The information processing apparatus described. Holding a message with an associated time,
Each of the held messages is divided into words, and a word sequence that matches a predetermined pattern is extracted from each message as a topic, and specified for each topic. Obtaining a score based on time, and
The step of obtaining the score has a difference between the designated time and the time associated with the message for each topic of each message as a denominator component, and the associated time is newer than the message. An information processing method, comprising: calculating a density of a topic using an expression having a number of messages including the topic as a component of a numerator; and obtaining a representative value of the density as a topic score for each topic. Holding a message with an associated time;
Dividing each held message into words, extracting a sequence of words that match a predetermined pattern from each segmented word sequence as a topic from each message, for each topic of each message Associating a time associated with the message and a number of messages that are newer than the message and that include the topic;
Holding the associated result;
Obtaining a score based on a specified time for a topic in the held associated result; and
The step of obtaining the score refers to the held associated result, and uses the difference between the designated time and the time associated with the message for each topic of each message as a denominator component. The associated time is newer than the message and the number of messages including the topic is used as an element of the numerator to obtain the density of the topic, and the representative value of the density for each topic is the topic score. An information processing method characterized by It has a step of storing topic information for each topic of each message in a maximum of N messages in ascending order of associated time, and the topic information is associated with the time associated with the topic and the message. Including the number of messages whose time is newer than the message and includes the topic,
When a message associated with a time is input, the input message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, For each extracted topic, the number of messages of topic information having a topic that matches the held topic is increased by 1, or the time associated with the input message is associated The topic information with a newer time than the message and the number of messages including the topic is created, added to the retained topic information, and the number of messages is stored in the index holding unit before the adding process. When topic information for N cases is held, the topic information generated from the message with the oldest associated time is deleted. And the floor,
Obtaining a score based on a specified time for a topic held;
In the step of obtaining the score, referring to the retained topic information, the difference between the time specified for each topic of each message and the time associated with the message is a denominator component, The density of the topic is obtained by using an expression having the number of messages whose associated time is newer than the message and includes the topic as a component of the numerator, and the representative value of the density is obtained as a topic score for each topic. An information processing method characterized by this. Holding a message with an associated time,
Each of the held messages is divided into words, and a word sequence that matches a predetermined pattern is extracted from each message as a topic, and specified for each topic. Having a stage for obtaining a score based on time,
In the step of obtaining the score, the difference between the designated time and the time associated with the message for each topic of each message has a denominator component, and for each word constituting the topic, The value of the topic is obtained by using an expression having the number of messages whose associated time is newer than the message and includes the word as a component of the numerator, and the representative value of the value is obtained as the topic score for each topic. An information processing method characterized by this. Holding a message with an associated time;
Dividing each held message into words, extracting a sequence of words that match a predetermined pattern from each segmented word sequence as a topic from each message, for each topic of each message Associating the time associated with the message and each word constituting the topic with the number of messages with the associated time newer than the message and including the word;
Holding the associated result;
Obtaining a score based on a specified time for the topic in the held result, and
In the step of obtaining the score, the stored associated result is referred to, and the difference between the designated time and the time associated with the message is used as a denominator component for each topic of each message. Then, for each word constituting the topic, the value of the topic is obtained using an expression having the number of messages including the word and the associated time is newer than the message. An information processing method characterized in that a representative value is obtained as a topic score. It has a step of storing topic information for each topic of each message in a maximum of N messages in ascending order of associated time, and the topic information constitutes the topic and the time associated with the message and the topic And the number of messages in which the associated time of each word is newer than the message and includes the word,
When a message associated with a time is input, the input message is divided into words, a word sequence that matches a predetermined pattern is extracted as a topic from the divided word sequence, For each word included in the input message, the message number of the word of the topic information having the retained word is increased by 1, and for each extracted topic, For each word constituting the topic and the time associated with the input message, topic information is created in which the associated time is newer than the message and the number of messages including the word is zero. , Added to the topic information, and when the topic information for the number N of messages is held in the index holding unit before the adding process, Comprising the steps of: time to remove the topic information generated from the oldest message that,
Obtaining a score based on a specified time for the topic of the topic information held,
The step of obtaining the score refers to the retained topic information, and has a difference between the designated time and the time associated with the message for each topic of each message as a denominator component, For each word that constitutes the topic, the value of the topic is determined using an expression having the associated number of messages that are newer than the message and the number of messages including the word as a component of the numerator. An information processing method characterized in that a representative value is obtained as a topic score.   A step of holding the number of topics that holds the number of messages in which each topic appears in the message to be processed, and the score of each topic is a component of the denominator of the number of messages for the topic in the topic appearance holding unit The information processing method according to any one of claims 9 to 11, wherein the information is obtained by weighting with an expression. The number of messages in which each word appears in the message to be processed is held, and the score of each topic is determined based on the time associated with each word constituting the topic and newer than the message. The information processing according to any one of claims 12 to 14, wherein the number of messages included is calculated by weighting with an expression having the number of messages for the word in the word appearance holding unit as a constituent element of a denominator. Method.

Images