Technique for searching for keywords determining event occurrence

• US 8,005,829 B2
• Filed: 03/07/2008
• Issued: 08/23/2011
• Est. Priority Date: 03/08/2007
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Takeuchi, Hironori, Roy, Shourya, Subramaniam, L. Venkata, Nasukawa, Tetsuya
• Primary Examiner(s): Ali; Mohammad
• Assistant Examiner(s): Tran; Bao G
• Application Number:
  US12/044,378
• Publication Number:
  US 20080256063A1
• Time in Patent Office: 1,264 Days
• Field of Search: 707/999.001, 707/999.104, 707/999.107, 707/730, 707/765, 700/245, 705/14.54
• US Class Current: 707/730
• CPC Class Codes: G06F 16/90344 by using string matching techniques
• CPC Class Codes: G06Q 30/0256 User search

Abstract

A keyword search system including a text input unit for inputting subtexts obtained by dividing each text into parts, while associating the subtexts with an event through a process recorded in the text; a prediction device adjuster for adjusting a corresponding event prediction device to maximize the percentage of text in which the inputted event is identical to a prediction result in a first text group selected from the subtexts; a prediction processor for generating a prediction result for each section, by inputting each text in a second text group selected from the corresponding subtexts in the adjusted event prediction device; and a search unit for calculating the prediction precision for the second text group of the event prediction device using a comparison between the inputted event and the prediction result for each subtext, and searching for keywords in sections with a certain degree of prediction precision.

Claims

1. A system for searching a plurality of texts of recorded conversations to determine keywords leading to an event occurrence in a plurality of texts wherein the texts each have a progress leading up to the event occurrence recorded therein, comprising:

a processing device for executing units;

a text input unit for inputting a plurality of subtexts while associating each of the subtexts with an event occurring through the process recorded in each of the texts, the subtexts obtained by selecting parts corresponding to each of a plurality of predetermined sections in each of the plurality of texts;

a plurality of event prediction devices corresponding to the plurality of sections respectively, the prediction devices each outputting a prediction result of an event occurring through the process recorded in the subtexts corresponding to the prediction device, the prediction result based on the appearance frequency of each word in the subtexts;

a prediction device adjuster for adjusting the event prediction device corresponding to each of the plurality of sections, so as to maximize the percentage of agreeing texts to a first text group selected from the plurality of corresponding subtexts, the agreeing texts each indicating the content in which the inputted event agrees with the prediction result;

a prediction processor for generating the prediction result for each of the plurality of sections, by selecting a second text group, which is different from the first text group, from the plurality of subtexts corresponding to the section, and then by inputting each text of the second text group in the adjusted event prediction device corresponding to the section; and

a search unit for calculating the prediction precision for the second text group of the event prediction device of each of the sections, the prediction precision based on a comparison result between the inputted event and the prediction result for each subtext, for searching for keywords in a section which have a higher degree of prediction precision than a predetermined reference value, and for outputting the keywords,

wherein the prediction device adjuster randomly selects the first text group from the plurality of subtexts corresponding to each of the plurality of sections, and

the prediction processor selects, as the second text group, some of the plurality of subtexts corresponding to each of the plurality of sections, by excluding the first text group from the subtexts.

2. The system according to claim 1, wherein

each of the plurality of texts is one in which the progress leading up to the event occurrence is recorded in chronological order,

the text input unit inputs records of a plurality of periods as the plurality of subtexts, the records selected from each of the texts, the records each corresponding to a period from the head of the text to one of a plurality of predetermined points of time, and

the search unit searches for keywords in a section where a first section is excluded from a second section including the first section, on condition that the calculated prediction precision for the second section is higher than that of the first section.

3. The system according to claim 2, wherein the search unit makes a search while setting, as the keywords, words having the number of appearances whose increasing rate of the second section to the first section is higher than the increasing rate of the text volume in the second section to that in the first section.

4. The system according to claim 2, wherein the search unit searches a plurality of texts corresponding to the section where the first section is excluded from the second section, while setting, as the keywords, words having the appearance frequencies that are different between texts corresponding to a first event and a second event to a larger extent than the other words.

5. The system according to claim 4, wherein the search unit calculates an evaluation value of each word, the evaluation value taking a larger value according to the increasing rate of the number of appearances in the second section to that in the first section and taking a larger value according to the difference between the frequencies, and makes a search while setting words with larger calculated evaluation values as the keywords.

6. The system according to claim 1, further comprising a display for displaying the subtexts corresponding to a section, while associating the subtexts with the keyword searched out by the search unit and identification information of the section searched for the keywords.

7. The system according to claim 4, further comprising:

a category input unit for receiving the input of category information which shows the category of each keyword searched out by the search unit; and

a decision tree generator for generating a decision tree for predicting an event using a text newly inputted, the decision tree including a node, an edge and a leaf node, the node indicating each piece of category information, the edge indicating whether or not a keyword corresponding to the category information appears in the text, and the leaf node indicating each event.

8. The system according to claim 1, the system facilitating a conversation leading to an occurrence of a desired event, the system further comprising:

a category input unit for receiving the input of category information which shows the category of each keyword searched out by the search unit from an already-inputted text;

a controller for sequentially inputting parts of a text in which a conversation in progress is recorded, and for generating at least one of an index value indicating the possibility of an occurrence of each event, and a keyword that should be spoken in order for a desired event to occur, the index value and keyword based on the frequency at which each of the keywords appears in the sequentially inputted text, the keywords already searched out as those to determine each event; and

a display for displaying, for a speaker of the conversation in progress, at least one of the index value indicating the possibility of an occurrence of each event and the keyword that should be spoken in order for the desired event to occur.

9. A method for causing a computer to search a plurality of texts of recorded conversations to determine keywords leading to an event occurrence, the texts each having a progress leading up to the event occurrence recorded therein, the method comprising the steps of:

inputting, for each of the plurality of texts, a plurality of subtexts which are obtained by selecting parts corresponding to a plurality of predetermined sections from the texts while associating the subtexts with events which occurred through the process recorded in the texts; and

when a plurality of event prediction devices are provided corresponding to the plurality of sections respectively, the prediction devices each outputting a prediction result of an event occurring through the process recorded in a corresponding one of the subtexts, the prediction result based on the appearance frequency of each word in the corresponding subtext,

adjusting the corresponding event prediction devices, for each of the plurality of sections, to maximize the percentage of agreeing texts to a first text group selected from the plurality of corresponding subtexts, the agreeing texts each indicating the content in which the inputted event agrees with the prediction result;

for each of the plurality of sections, selecting a second text group, which is different from the first text group, from the plurality of subtexts corresponding to the section, and then by inputting each text of the second text group in the adjusted event prediction device corresponding to the section, and generating the prediction result; and

calculating the prediction precision for the second text group of the event prediction device of each of the sections, the prediction precision based on a comparison result between the inputted event and the prediction result for each subtext, for searching for keywords in a section which have a higher degree of prediction precision than a predetermined reference value, and for outputting the keywords,

and wherein the adjusting comprises randomly selecting the first text group from the plurality of subtexts corresponding to each of the plurality of sections, and

wherein the selecting of the second text group comprises selecting the plurality of subtexts corresponding to each of the plurality of sections, by excluding the first text group from the subtexts.

10. A program storage device storing a program for causing a computer to function as a system for searching a plurality of texts of recorded conversations for determining keywords that lead to an event occurrence, the texts each having a progress leading up to the event occurrence recorded therein, the program causing the computer to function as:

a text input unit for inputting a plurality of subtexts while associating each of the subtexts with an event occurring through the process recorded in each of the texts, the subtexts obtained by selecting parts corresponding to each of a plurality of predetermined sections in each of the plurality of texts;

an event prediction device provided corresponding to the plurality of sections respectively, the prediction devices each outputting a prediction result of an event occurring through the process recorded in the subtexts corresponding to the prediction device, the prediction result based on the appearance frequency of each word in the subtexts;

a prediction device adjuster for adjusting the event prediction device corresponding to each of the plurality of sections, so as to maximize the percentage of agreeing texts to a first text group selected from the plurality of corresponding subtexts, the agreeing texts each indicating the content in which the inputted event agrees with the prediction result;

a plurality of prediction processors for generating the prediction result for each of the plurality of sections, by selecting a second text group, which is different from the first text group, from the plurality of subtexts corresponding to the section, and then by inputting each text of the second text group in the adjusted event prediction device corresponding to the section; and

a search unit for calculating the prediction precision for the second text group of the event prediction device of each of the sections, the prediction precision based on a comparison result between the inputted event and the prediction result for each subtext, for searching for keywords in a section which have a higher degree of prediction precision than a predetermined reference value, and for outputting the keywords

wherein the prediction device adjuster randomly selects the first text group from the plurality of subtexts corresponding to each of the plurality of sections, and

the prediction processor selects, as the second text group, some of the plurality of subtexts corresponding to each of the plurality of sections, by excluding the first text group from the subtexts.