JP2004133771A - Device, method, and program for question and answer - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a question and answer device, etc., which has high accuracy and by which a user easily looks at an answer. <P>SOLUTION: A question sentence in a natural language inputted by a user is subjected to a morphological analysis, etc., to extracts a word. A second word and an interrogative included in the extracted word are used to estimate an attribute with which an answer coincides with the question sentence should be provided. For example, when the question sentence is "What is the population of Afghanistan ?", an attribute with which the answer should be provided is "○○ people (a numeral plus a cardinal number)". Then, a sentence including the extracted word and a word coinciding with the attribute of the answer is retrieved, and words provided with the attribute are extracted from the retrieved sentence as an answer candidate. Furthermore, the extracted answer candidates are accumulated on the basis of the number of overlapping times and preparing time and arranged and displayed in the order of the number of overlapping times and the preparing time. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a question answering device and the like, for example, to a device that outputs an answer to an input question.
[0002]
[Prior art]
Conventionally, when there is a question item and an answer to the question item is obtained by using a computer, the user generally inputs a question sentence to the computer and searches for a document from which an answer is likely to be obtained.
In this case, the computer extracts a keyword from the question sentence input by the user, and searches for a document from the document set using the keyword. Then, the user reads the searched document and obtains a necessary answer.
[0003]
By the way, most of such information retrieval systems are “document” retrieval systems, and the user has to read a huge amount of retrieved results and extract necessary information.
Further, even if the result is read with difficulty, the information does not always include the information that the user wants to know.
For example, if the user inputs "When did the stadium open?" As the question text, the information that the user wants to know is the date when the stadium was opened, but the keyword obtained from the question text ○ "○ X stadium" , "Open" search results may not include the date of opening.
[0004]
As in the above example, there is a case where an intended answer cannot be obtained only by a keyword extracted from a question sentence, so a question answering system that analyzes a question sentence and returns an answer has been devised.
The question answering system is a system that reads a search intention as to what kind of information the user wants from a question sentence, and extracts a specific part that matches the content of the user's question. The operation of the question answering system is briefly described as follows.
First, the input question text is analyzed to extract question information, and a part of the document (called a passage) that matches the question text is obtained.
Then, an answer of the question sentence is searched from the analysis result, and an appropriate answer is returned.
[0005]
[Problems to be solved by the invention]
By using this question answering system, answers can be narrowed down to some extent. However, an answer matching the question was not always obtained. Furthermore, when there are a plurality of answers or the answers are fluid such as numerical values, these answers have not been narrowed down and displayed simply.
[0006]
Therefore, an object of the present invention is to provide a question answering device with high accuracy and easy to see answers.
[0007]
[Means for Solving the Problems]
In order to achieve the object, according to the first aspect of the present invention, a question sentence acquiring unit for acquiring a question sentence, a word extracting unit for extracting a word from the acquired question sentence, Estimating means for estimating an attribute to be included in an answer to the question sentence using question words included in the word; and matching the question sentence using at least one of the extracted words and the estimated attribute. Search means for searching for a document including the answer candidate to be answered, answer generating means for summing up the answer candidates included in the searched document to generate an answer to the question sentence, and answer output means for outputting the generated answer And a question answering device characterized by comprising:
Here, in the question answering apparatus according to claim 1, when the word extracted by the word extracting means does not include a question word, the estimating means uses a word at the end of the question sentence, It is also possible to configure so as to estimate an attribute to be provided in an answer to the question sentence.
The invention according to claim 2 is characterized in that the answer generation means sums up the number of answer candidates in the searched document and generates an answer by arranging the answer candidates in descending order of the number of answers. The question answering device according to claim 1 is provided.
In the invention according to claim 3, when the duplication number of the largest number of answer candidates reaches a predetermined ratio of the number of answer candidates included in the searched document, the answer generation unit determines 3. The question answering device according to claim 2, wherein an answer is generated only from a plurality of large answer candidates.
In the invention according to claim 4, the answer generation unit acquires each creation time of the duplicate answer candidate, sets the most frequent time as the creation time of the duplicate answer candidate, and 3. The question answering apparatus according to claim 2, wherein answers are generated by arranging the answers in the order of the creation time.
In the invention according to claim 5, when there are a plurality of answer candidates included in the searched document, the answer generating unit generates the answer candidates included in the searched document. The question answering apparatus according to claim 1, wherein the answer is generated by arranging the answers in order of date and time.
According to the invention described in claim 6, in a computer including a question sentence acquisition unit, a word extraction unit, an estimation unit, a search unit, an answer generation unit, and an answer output unit, A question sentence obtaining step of obtaining a question sentence, a word extracting step of extracting a word from the obtained question sentence by the word extracting means, and the estimating means using question words included in the extracted word. An estimating step of estimating an attribute to be included in an answer to the question sentence, and using the at least one of the extracted words and the estimated attribute, the search means includes an answer candidate matching the question sentence. A search step of searching for a document; and an answer generation step of, by the answer generation means, summing up answer candidates included in the searched document and generating an answer to the question sentence. Provides a question answering wherein the configured answer that the generated from, and answer output step of outputting by the answer output unit.
In the invention according to claim 7, a question sentence acquisition function of acquiring a question sentence, a word extraction function of extracting a word from the acquired question sentence, and a question word included in the extracted word are used. An estimation function for estimating an attribute to be included in an answer to the question sentence, and a search function for searching for a document including an answer candidate matching the question sentence, using at least one of the extracted words and the estimated attribute. A question answering program for realizing, by a computer, an answer generating function for summarizing answer candidates included in the searched document and generating an answer to the question sentence, and an answer output function for outputting the generated answer I will provide a.
In addition, a question sentence obtaining function for obtaining a question sentence, a word extracting function for extracting a word from the obtained question sentence, and an answer to the question sentence are provided using question words contained in the extracted word. An estimation function for estimating a power attribute; a search function for searching for a document including an answer candidate that matches the question sentence using at least one of the extracted words and the estimated attribute; The answer generation function for generating the answer to the question sentence, and the answer output function for outputting the generated answer, and a computer-readable program storing a question program for realizing the answer by the computer. A storage medium may be provided.
[0008]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings.
(1) Overview of the embodiment
The question answering apparatus 1 (FIG. 1) of the present embodiment acquires a question sentence in a natural language created by a user (question input means 3), analyzes this, and extracts words (question analysis means 5). If the extracted word includes a question word, an attribute (answer type) to be provided in an answer matching the question sentence is estimated from the extracted other words and this question word.
For example, if the question sentence is "What is the population of Afghanistan?", From the words "Afghanistan", "Population", and the question word "How many", the answer is "XX people (numerals + classifiers)" The attribute of the answer is estimated, for example, at the part of speech level so as to provide
[0009]
Then, the question answering apparatus 1 searches the document database 11 for an answer candidate that matches the extracted word and attribute, and extracts an answer candidate from the searched document data (document search means 9).
Next, if there are a plurality of the same answer candidates among the extracted answer candidates, the number of duplicates is counted, or the time of creation of the document data from which the answer candidates are extracted is checked, and the extracted answer candidates are totaled. (Aggregating means 13).
[0010]
Finally, the extracted answer candidates are arranged and output by a predetermined method (answer output means 15), and the answer is presented in a form that is easy for the user to see.
There are two types of answer display modes: an overlap order display in which answers are arranged in descending order, and a creation time order display in which creation answers are displayed in order of creation time. The user can select a display style.
[0011]
(2) Details of the embodiment
FIG. 1 is a diagram illustrating a functional configuration of a question answering apparatus 1 according to the present embodiment.
The question answering apparatus 1 is composed of components such as a question input unit 3, a question analysis unit 5, an answer estimation unit 7, a document search unit 9, a document database 11, a tally unit 13, an answer output unit 15, and the like.
As will be described later, these components are realized by software by executing a question answering program on a computer.
[0012]
The question input unit 3 receives an input of a question sentence (query) from a user. The input of the question sentence is performed, for example, by inputting into the question sentence input field displayed on the display of the question answering apparatus 1 using a keyboard or the like.
The question sentence input by the question input means 3 is a 5W1H type question sentence composed of a natural language such as "Who is the 35th President of the United States?" .
The question analysis unit 5 extracts a word (term) from the question sentence input from the question input unit 3 and determines the presence or absence of a question word. In a language such as Japanese that is not separated, words are extracted by performing language analysis such as morphological analysis and syntax analysis.
[0013]
The answer estimating means 7 estimates attributes to be included in the answer using the words extracted by the question analyzing means 5.
The method of estimating the attribute of the answer differs depending on whether or not the word extracted from the question sentence has the question word (5W1H). When there is a question word in the question sentence, the attribute of the answer is estimated using the extracted other words and the question word.
For example, suppose that there is a question sentence "When was the stadium opened?"
The question answering apparatus 1 estimates that the answer type is date and time from the question word “when” included in the question sentence, and the answer is targeted for an attribute including “year, month, day, time” as an attribute. Then, when the retrieved document includes a word that matches the attribute estimated as “○ / ○ / day (numerical number + classifier)”, this word is an answer candidate.
If the question is "Who is the president of XX Corporation?", The attribute that should be included in the answer from the word "President" and the question word "Who" is a unique person name.
Further, when the question sentence is "Where is the country with the largest population?", The attribute that should be included in the answer is the country name from the word "Country" and the question word "Where".
[0014]
If the question is omitted, the answer type is determined from the last word of the question (if any, the part of speech of the word). For example, suppose there is a question sentence "What is the name of the president of XX Automobile?"
The last word of the question sentence in this case is “name”. From "President" (person attribute) immediately before "Name", the attribute to be provided in the answer is a unique personal name.
If the question is "What is the oldest temple?", The name of the temple will be the answer candidate from the word "Tera" at the end.
[0015]
The document search means 9 removes unnecessary stop words (for example, question words) from the words extracted by the question analysis means 5. Then, a document is searched in the document database 11 using these words and the attributes of the answer estimated by the answer estimating means 7. As a search method, a vector space method, a Boolean search, or the like is used. The document database 11 is a database in which various types of document data from which answers are created are stored.
As described above, by including the attribute of the answer candidate in the search condition, the document to be searched includes the answer requested by the user. Thereby, the accuracy of the answer is improved.
[0016]
Alternatively, after a document is searched using the word extracted by the question analysis unit 5, a document that does not include an answer candidate that matches the attribute estimated by the answer estimation unit 7 may be removed.
In this case, since the searched document is filtered by using a predetermined answer candidate, the document output from the document search unit 9 includes the answer requested by the user.
[0017]
Here, a specific processing example of a question sentence including the 5W1H type will be described.
It is assumed that a question sentence “When did the XX stadium open?” Is input from the question input means 3.
The question analysis means 5 extracts words such as “○ × ballpark”, “when”, “opening” from the question text. Then, the answer estimating means 7 checks whether or not the extracted word includes a question word. In this case, the question sentence includes the question word “when”. Therefore, the answer estimating means 7 estimates that the answer type is a word indicating the date and time, and determines that the answer candidate is a word having an attribute of “○ month ○ day (numerical number + classifier)”.
Next, the document retrieval unit 9 extracts words such as “xx ballpark” and “opening business” extracted by the question analysis unit 5 and attributes “xx month and day (numerical number + classifier)” estimated by the answer estimation unit 7. Search for documents containing words with.
[0018]
Next, a specific processing example of a question sentence in which the 5W1H type is omitted will be described.
It is assumed that a question sentence “What is the price of game machine ○ △?” Is input from the question input means 3. The question analysis means 5 extracts words such as “game machine ○ △” and “price” from the question sentence. Then, the answer estimating means 7 confirms whether or not the extracted word includes a question word. In the case of this question, the question is omitted because the question is omitted.
[0019]
Therefore, the answer estimating means 7 estimates that the answer type is a word indicating the amount of money from the word "price" at the end of the question sentence, and the answer candidate has the attribute "XX circle (numeral + particle)". Judge that it is a word.
Next, the document search unit 9 includes words such as “game machine △” and “price” extracted by the question analysis unit 5 and answer candidates having an attribute of “XX yen (numeral + particle)”. Search for documents.
[0020]
Further, as another example of the question sentence in which the 5W1H type is omitted, "What is the population of Afghanistan?"
In this case, the answer estimating means 7 estimates that the answer type is a word indicating the number of persons based on the "population" included in the question sentence, and the answer candidate has an attribute of "XX person (numeral + particle)". Judge that it is a word. Then, the document search means 9 searches for a document including a word having the attribute of “Afghanistan”, “population”, and “XX (numerals + particles)”.
[0021]
Next, the counting means 13 acquires answers matching the answer candidates from the documents searched as described above, and counts these.
In the present embodiment, two types of tallying methods are provided, namely, tallying in ascending order of response and tallying in the order of time when the document was created, so that the user can select either one. .
The answer output unit 15 outputs a tally answer using the tally result. The output total answer is presented to the user, for example, displayed on a display or printed.
[0022]
Regarding the method of totaling the answers in descending order, if only one answer result is returned from the search results, only that answer is displayed, and the rest is not displayed. This is due to the judgment that the majority of answers are close to the correct answer. In the present embodiment, only the first-ranked answer is output when the number of totals in the first place becomes half or more of the total number of totals. Here, the value of 1/2 is an example, and is a value that can be arbitrarily changed.
[0023]
For example, as a result of the tabulation means 13 totaling the answers to the question sentence “When did the XX stadium open?”, 90 answers for “March 31” and 5 answers for “March 25” were answered. It is assumed that "April 29" is five.
In this case, the answer “March 31” accounts for more than half of the total number of answers (in the case of the example, 90/100 of the total number of answers). Therefore, the answer output unit 15 outputs the answer “March 31” as the correct answer. The other answers “March 25” and the answer “April 29” are not output.
[0024]
When a plurality of answers are returned in a dispersed manner without concentrating on one answer result, all of these answers are output. In the present embodiment, all answers are output when the number of totals in the first place is less than half of the total number of totals. Note that the value of 1/2 is an example and can be arbitrarily changed.
[0025]
For example, as a result of tabulating the answers to the question sentence “What is the population of Afghanistan?”, 40 answers “19,000,000”, 30 answers “18,000,000”, and answers “18,879, Suppose that there were 30 “000 people”. In this case, since the number one answer is not more than half of the number of answers, all the answers are output.
As described above, since the question answering apparatus 1 can display the answers to the question sentences in descending order, it can be used for questionnaire analysis, for example.
[0026]
On the other hand, when the documents are totaled in the order of time when the documents were created, the counting means 13 acquires the date and time information when the document from which the answer was acquired was created. At this time, the date and time information that is the most frequent among the answers is set as the date and time information of the answer. The results can be displayed in chronological order by rearranging the results in chronological order or chronological order.
It is assumed that the user can select the sorting time setting from year, month, and day as an option.
[0027]
For example, as a result of summarizing the answers to the question sentence “What is the population of Afghanistan?”, Answer candidates “19,000,000”, “18,000,000”, “18,879,000” are obtained, It is assumed that the creation year of the document from which each answer candidate was obtained was 1999, 1996, and 1990, respectively.
In this case, the answer output unit 15 determines that the answer candidates are “19,000,000” (1999), “18,000,000” (1996), and “18,879,000” (1990). Is output in the order of and with the creation time added.
[0028]
The transition of the population of Afghanistan is displayed on the display of the question answering apparatus 1 together with the date of creation of the document.
As described above, the question answering apparatus 1 can display the result of the question sentence in the order of the creation time of the document from which the answer is obtained, so that the time transition of data, for example, the transition of statistics such as population can be viewed. Can be.
[0029]
FIG. 2 is a flowchart for explaining the operation of the question answering apparatus 1. The operation of each component described below is performed when a CPU (Central Processing Unit) mounted on the question answering apparatus 1 operates according to a question answering program.
First, the question input unit 3 acquires a question sentence input by the user (step 5). This question sentence is a question sentence in a natural language that is input by a user using a keyboard or the like in a question input field displayed on the display of the question answering apparatus 1.
Next, the question analysis means 5 obtains a question sentence from the question input means 3 and extracts words by morphological analysis or the like (step 10).
[0030]
Next, the answer estimating means 7 confirms whether or not the word extracted in step 10 includes an interrogative word (step 15). If a question word is included (Step 15; Y), the answer estimating means 7 determines an attribute to be included in an answer candidate matching the question sentence from the question word and other extracted words (Step 20). ).
If no question word is included (Step 15; N), the answer estimating means 7 determines an attribute to be included in the answer candidate matching the question sentence from the last word of the question sentence (Step 25).
[0031]
Next, the document search means 9 searches the document database 11 for a document including the word (excluding the question words) extracted in step 10 and the answer candidate having the attribute determined in step 20 or 25 (step 30). ). It is not always necessary to use all of the words extracted in step 10 in the search. For example, syntax analysis and semantic analysis are performed on a question sentence, and words to be used in the search are selected according to the results. May be configured.
[0032]
Next, the counting means 13 extracts answer candidates from the searched documents (step 35), and counts the extracted answer candidates (step 40).
Among these processes, the extraction of the answer candidates in step 35 is performed by extracting words having the attribute determined by the answer estimating means 7 from the retrieved documents. For example, if the attribute to be included in the answer is “XX yen (numeral + particle)”, a word that matches this attribute, such as “300 yen”, is extracted from the document.
[0033]
The counting of the answer candidates in step 40 includes counting the number of extracted answer candidates, and adding information on the creation time to the answer candidates from the creation time of the document from which the answer candidates are extracted.
For example, for the answer candidate “20 million”, there are 100 cases with a creation time of 1999 (here, it is assumed that the user has selected the year as the creation time) and a creation time of 1998 If there are five cases, 1999, which occupies a large number of times, is assigned to this answer candidate.
[0034]
Next, the answer output means 15 specifies the answer display mode based on the information input by the user (the user can select the display mode in the order of the number of duplicates or in the order of the creation time) (step 45). ), And outputs the answer so that the answer is displayed in the specified display form (steps 50 and 60).
The display forms that can be selected by the user include displaying answer candidates in descending order of the number of duplicates and displaying them as answers (duplicate order display) and displaying the answer candidates in the order of creation time assigned to the answer candidates. There is a form of answer (display in order of creation time).
When the user selects the overlapping display, the answer output unit 15 outputs the answer so that the answer candidates are displayed in descending order of the number of answer candidates (step 50), and the user selects the display in the order of creation time. In this case, the answer is output so that the answer candidates are displayed in chronological order (step 60).
[0035]
In the above-described order of the creation time, the overlapping answer candidates are combined into one, and one creation time is given to the one group of the combined answer candidates. It is not limited to. For example, answer candidates may be divided into groups having the same creation time, and further, overlapping answer candidates in each group may be collected.
Further, the user may be able to view the document including the answer from the answer result.
[0036]
FIG. 3 is a diagram illustrating an example of a hardware configuration of the question answering apparatus 1.
The question answering apparatus 1 is configured using, for example, a personal computer, and a ROM (Read Only Memory) 70, a RAM (Random Access Memory) 72, an input device 74, an output device 78, Peripheral devices such as a communication control device 82, a storage device 88, a storage medium driving device 76, and an input / output I / F (interface) 84 are connected to each other.
The bus line 73 mediates transmission and reception of control signals and data signals transmitted and received between the CPU 68 and peripheral devices.
[0037]
The CPU 68 performs various types of information processing and control, such as generating and displaying an answer to a question sentence according to a question answering program 50 described later, controlling the entire question answering apparatus 1, and the like.
The ROM 70 is a read-only storage device that stores various programs, data, parameters, and the like for the CPU 68 to perform various calculations and controls. The CPU 68 can read programs, data, parameters, and the like from the ROM 70, but cannot rewrite or delete them.
[0038]
The RAM 72 is a readable and writable storage device used as a working memory by the CPU 68. The CPU 68 can write and erase programs and data in the RAM 72. In the present embodiment, the RAM 72 receives and analyzes the input of a question sentence, searches for a document including an answer candidate, generates and displays an answer, and performs other various information processing. Area can be secured.
[0039]
The input device 74 is configured by an input device such as a keyboard and a mouse.
The keyboard is a device for inputting information such as characters and numbers to the question answering device 1.
The keyboard includes keys for inputting kana and English characters, numeric keys for inputting numbers, various function keys, cursor keys, and other keys.
The user can input a question sentence to the question answering apparatus 1 using a keyboard in a natural language.
[0040]
The mouse is a pointing device. When operating the question answering apparatus 1 using a GUI (Graphical User Interface) or the like, predetermined information can be input by clicking a button or icon displayed on the display device with a mouse.
For example, it is possible to display an overlap order display button and a creation time order display button on a display, and to select a display mode by clicking a button with a mouse operation by a user.
[0041]
The output device 78 includes, for example, a display device, a printing device, and the like.
The display device is a device for presenting, on a screen, information constituted by a display such as a CRT (Cathode Ray Tube) display, a liquid crystal display, and a plasma display.
The display device can display a question input field for inputting a question sentence, a generated answer, and the like.
[0042]
The printing device is a device that prints on a print medium such as paper in the form of a text such as a tally answer. The printing device is configured by various printer devices such as an ink jet printer, a laser printer, a thermal transfer printer, and a dot printer. By using the printing device, the answer can be printed on a print medium. When printing the answer on the print medium, the answer is printed in a form selected by the user between the overlapping order display and the creation time order display.
[0043]
The communication control device 82 is a device for connecting the question answering device 1 to a server device or another terminal device via a communication line, and is composed of a modem, a terminal adapter, and other devices.
When the communication control device 82 is used, it is connected to, for example, the Internet or a LAN (Local Area Network), and it is also possible to search for document data stored in another terminal device or server device connected to these networks. is there.
The communication control device 82 is controlled by the CPU 68 and transmits and receives signals and data to and from these terminal devices and server devices according to a predetermined protocol.
[0044]
The storage medium drive 76 is a drive for reading and writing data by driving a removable storage medium. The removable storage medium, e.g., magneto-optical disks, magnetic disks, magnetic tape, semiconductor memory, paper tape data was punched, and the like CD-ROM. The CD-ROM and the paper tape can only be read.
[0045]
The storage device 88 includes a readable and writable storage medium, and a drive device for reading and writing programs and data from and to the storage medium. Although a hard disk is mainly used as the storage medium, other storage mediums such as a magneto-optical disk, a magnetic disk, and a semiconductor memory, for example, such as a readable / writable storage medium, may be used.
[0046]
The storage device 88 stores a question answering program 50, the document database 11, other programs 52, other data 54, and the like.
The CPU 68 can read and write programs and data from and to the storage device 88 by driving the driving device of the storage device 88.
[0047]
The question answering program 50 is a program for causing the CPU 68 to perform a question answering function. When the question answering program 50 is read and executed by the CPU 68, each component of the question inputting means 3 to the answer outputting means 15 (excluding the document database 11) shown in FIG.
The document database 11 stores document data from which answer candidates are obtained, for example, as text data.
[0048]
The other programs 52 control, for example, the communication control device 82 to maintain communication with the terminal device or the server device connected to the question answering device 1 via a network, or to answer questions such as memory management and input / output management. It is configured by various programs such as an OS (Operating System) which is basic software for operating the device 1.
The other data 54 includes, for example, a morphological dictionary for the question analysis unit 5 to perform morphological analysis of the question sentence, and a syntactic analysis dictionary and a semantic analysis dictionary to be used for further parsing and semantic analysis. Dictionaries and other data are stored.
[0049]
The input / output I / F 84 includes, for example, a serial interface or an interface of another standard. By connecting an external device corresponding to the interface to the input / output I / F 84, the function of the question answering apparatus 1 can be expanded. Examples of such external devices include a storage device such as a hard disk, a speaker, and a microphone.
[0050]
In the present embodiment, the question answering apparatus 1 is configured as a stand-alone using a personal computer. However, the present invention is not limited to this. The question answering apparatus 1 may be constructed as a question answering server that provides a question answering service on a network.
The configuration of the question answering server is basically the same as that of the question answering device 1 shown in FIG. When a question sentence receiving means for receiving a question sentence from the user terminal using the communication control device 82 and an answer transmission means for transmitting the generated answer to the user terminal using the communication control device 82 are provided, Can function as a server.
[0051]
Next, how a question sentence is processed by hardware will be described with reference to FIGS. 4 to 9.
FIG. 4 is a diagram showing a question input screen displayed on the display of the question answering apparatus 1. As shown in the figure, on the question input screen, a question input field for inputting a question message is displayed under an instruction of “Please input a question message”. The user uses a keyboard or the like to enter a question such as "What is the population of Tokyo?"
[0052]
Further, a radio button for selecting a display mode is provided below the question input column. By clicking a radio button with a mouse or the like, it is possible to select either the overlapping order display or the creation time order display. When one of the radio buttons is selected, the selected state of the already selected radio button is released. In the figure, the overlapping order display is selected.
An execution button is provided at the lower part of the question input screen. After inputting a question sentence in the question input field, selecting a display mode, and clicking this execution button with a mouse operation, the CPU 68 generates an answer. Start.
[0053]
FIG. 5 shows that the document search means 9 retrieves the document database 11 from the words "Tokyo" and "population" extracted from the question sentence and the attribute "XX person (numeral + classifier)" of the answer candidate determined from the question words. FIG. 9 is a diagram showing an example of a logical configuration of data after sorting by using the search result, which is sorted by the totaling unit 13. The rearranged word data is stored in the RAM 72.
The retrieved document data is organized into a creation time, an answer candidate, and document data using the document ID as a key. The document ID is a number, a symbol, or the like assigned to each document data in order to uniquely identify each document data. The creation time is acquired in units of year, month, day, hour, minute, and second.
The answer candidate is obtained by extracting a word that matches the attribute that the answer candidate should have from the document data.
[0054]
FIG. 6 is a diagram showing, by using a bar graph, a tabulated result obtained by tabulating the data after the sorting in FIG. The result of the aggregation is stored in the RAM 72.
The vertical axis of the graph represents each answer candidate, and the horizontal axis represents the duplication number of each answer candidate. The number shown at the right end of the bar graph represents the creation time given to the answer candidate.
FIG. 7 shows an example of an answer display screen in which answer candidates are arranged and displayed in the overlapping display order.
[0055]
On the other hand, FIG. 8 shows an example of an answer display screen in which answer candidates are arranged and displayed in order of creation time. When displaying answer candidates in order of creation time, a graph display button is provided in the screen, and when this button is clicked with a mouse operation, the time transition of the answer candidates can be displayed in a graph.
FIG. 9 is a diagram showing an example of a graph display screen displayed when the graph display button is clicked on the answer display screen of FIG. The transition of the population of Tokyo can be visually presented by a graphical display.
As described above, the graph display screen is particularly useful when it is desired to know a temporal change of a target, such as a change in consumption of crude oil and a change in economic growth rate.
[0056]
In the question answering apparatus 1 described above, the following effects can be obtained.
When a document is searched, an answer candidate that matches the question message is set as a search condition. Therefore, the searched document includes an answer candidate for the question message. Therefore, the accuracy of the answer is improved.
Also, answer candidates extracted from the retrieved documents are totaled and presented in a form that is easy for the user to see. Specifically, the answer candidates can be arranged and displayed in descending order of the duplication number, or the answer candidates can be provided with creation time information and displayed in order of the creation time.
Further, the question answering apparatus 1 of the present embodiment can be mounted on a mobile device such as a mobile phone. In general, the amount of information to be displayed increases when a large number of items satisfying the conditions are found in a document search or the like. For example, a portable terminal with a small screen may lack the listability. However, by mounting the question answering apparatus 1, it is possible to display narrowed-down answer candidates or to display the data as visualized data such as a graph. Can be.
[0057]
As described above, one embodiment of the present invention has been described, but the present invention is not limited to the above-described embodiment, and various modifications can be made within the scope described in each claim.
[0058]
【The invention's effect】
According to the present invention, it is possible to provide a question answering device or the like with high accuracy and easy to see answers.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating a functional configuration of a question answering apparatus according to an embodiment.
FIG. 2 is a flowchart for explaining the operation of the question answering apparatus.
FIG. 3 is a diagram illustrating an example of a hardware configuration of a question answering apparatus.
FIG. 4 is a diagram showing an example of a question input screen.
FIG. 5 is a diagram showing an example of a logical configuration of the data after the sorting by the counting means.
FIG. 6 is a bar graph showing the result of totalizing the arranged information.
FIG. 7 is a diagram showing an example of an answer display screen in a multiple-number order display.
FIG. 8 is a diagram showing an example of an answer display screen in a creation time order display.
FIG. 9 is a diagram showing an example of a graph display screen in which answers in a creation time display are graphed.
[Explanation of symbols]
1 Question answering device
3 Question input means
5 Question analysis means
7 Answer estimation means
9 Document search means
11 Document database
13 Aggregation means
15 Answer output means
50 Question Answering Program
52 Other programs
54 Other data
68 CPU
70 ROM
72 RAM
73 bus line
74 input device
76 Storage medium drive
78 Output device
82 Communication control device
84 Input I / F
88 storage device

Claims (7)

A question sentence acquiring means for acquiring a question sentence,
Word extraction means for extracting a word from the obtained question sentence,
Estimating means for estimating an attribute to be provided in an answer to the question sentence, using a question contained in the extracted word,
Using at least one of the extracted words and the estimated attribute, searching means for searching for a document including an answer candidate matching the question sentence;
Answer generation means for summing up the answer candidates included in the searched document and generating an answer to the question sentence,
Answer output means for outputting the generated answer,
A question answering device characterized by comprising: 2. The question according to claim 1, wherein the answer generation unit sums up a plurality of answer candidates in the searched document, and generates an answer by arranging the answer candidates in descending order of the plurality of answer candidates. 3. Answering device. When the duplication number of the largest number of answer candidates reaches a predetermined ratio of the number of answer candidates included in the searched document, the answer generation unit may select an answer from only the largest answer candidate having the largest number of overlaps. The question answering device according to claim 2, wherein the question answering device is generated. The answer generation unit obtains each creation time of the overlapped answer candidate, sets the most frequent time as the creation time of the overlapped answer candidate, and arranges the overlapped answer candidates in the order of the creation time. The question answering device according to claim 2, wherein When there are a plurality of answer candidates included in the searched document, the answer generating unit generates an answer by arranging the answer candidates included in the searched document in order of the date and time when the answer candidate was created. The question answering apparatus according to claim 1, wherein In a computer including a question sentence acquisition unit, a word extraction unit, an estimation unit, a search unit, an answer generation unit, and an answer output unit,
A question sentence obtaining step of obtaining a question sentence by the question sentence obtaining means,
A word extraction step of extracting a word by the word extraction means from the acquired question text;
An estimating step of estimating an attribute to be provided in an answer to the question sentence by the estimating means, using a question word included in the extracted word;
A search step of searching, using at least one of the extracted words and the estimated attribute, a document including an answer candidate matching the question sentence by the search means;
An answer generating step of, in the answer generating means, summing up answer candidates included in the searched document and generating an answer to the question sentence;
An answer output step of outputting the generated answer by the answer output means,
A question answering method characterized by comprising: A question sentence acquisition function for acquiring a question sentence,
A word extraction function for extracting a word from the acquired question sentence,
Using an interrogative included in the extracted word, an estimation function for estimating an attribute to be provided to an answer to the question sentence,
Using at least one of the extracted words and the estimated attribute, a search function of searching for a document including an answer candidate that matches the question sentence;
An answer generation function that aggregates answer candidates included in the searched document and generates an answer to the question sentence;
An answer output function for outputting the generated answer,
Question answering program for realizing on a computer.

Images