JP2005275601A - Information retrieval system by voice - Google Patents

Abstract

Provided is a technique that enables a search for presenting a search result corresponding to a request by grasping a user's request that does not appear in a search item from the utterance content of the user.
A speech information retrieval system includes a retrieval item extraction unit that extracts a recognition word that is a keyword for a retrieval item from speech data including a utterance content of a user who desires information retrieval, and a user's speech from the speech data. A voice information extraction unit that extracts emotion information related to emotion; and a search control unit that generates and outputs a search expression including a search condition for at least one search item from the recognized word and the emotion information.
[Selection] Figure 1

Description

  The present invention relates to a voice response service field in which a search condition desired by a user is input by using a voice response such as a voice portal and information is provided.

  In recent years, technological development in speech response related fields such as speech recognition, speech synthesis, and speech dialogue has been remarkable. The voice response service is a service in which the system provides various services to the user through a dialogue between the system having the voice-related technology and the user. A typical voice response service is a voice portal that allows a user to access the system using voice from a personal computer or telephone and receive the service.

For example, in the case of searching a Web site with an Internet search engine using WWW (World Wide Web) or the like, the following methods have been taken.
(1) Search and display only data satisfying all items of the search condition.
(2) With respect to the search condition, a preset range is validated and a search is performed including data within the range.

Also, as a method of presenting the evaluation results, a method is generally used in which the search conditions are weighted and evaluated, comprehensive points are provided for the search results, and items (items) searched in descending order of the comprehensive points are presented. (For example, Patent Document 1).

  Also in a conventional voice response system, a method used in a conventional search engine such as WEB has been generally used.

In addition, as a technique including means for removing difficulty in searching unique to voice, for example, a technique for changing priority among items using an attached word obtained at the time of voice response (for example, Patent Document 2), and There is a technique (for example, Patent Document 3) that extracts information for efficient information retrieval from surrounding situations other than the user's utterance.
Japanese Patent Laid-Open No. 03-118662 Japanese Patent Laid-Open No. 01-263823 JP 2000-20090 A

  In Web search or the like, a user can set search conditions to be set at a time while viewing the search conditions. For this reason, it is possible to easily set, add, and change search conditions. In addition, setting of weighting between items, setting of search condition range, setting of one item and two conditions can be performed relatively easily.

However, in the voice response system, the search condition is set by voice dialogue. For this reason, the following voice-specific difficulties arise regarding the search conditions.
(1) The search condition cannot be visually confirmed. Also, fewer items can be set at a time than Web search. (2) It is difficult to set fine search conditions such as weighting, range setting, and one item / two conditions. (3) Since confirmation is performed by reading out setting items, it takes time to set.

In particular, in the search by the conventional voice response system, the search condition according to the request from the user (
Search keywords themselves), relaxation conditions for predetermined search conditions (standards for determining the range to hit the search target data as matching the search conditions), and weighting between predetermined search items (emphasis on items) The search was performed on the basis of the ranking. As described above, since the predetermined relaxation conditions and weighting are used, if the user's intention does not sufficiently match the relaxation conditions and weighting, the user may not be able to obtain a desired search result. there were. In this case, it is necessary to set search conditions again, and it may take time for the user to obtain a desired search result.

  In order to solve the above-described problems, an object of the present invention is to search information for grasping a user's request from the content of the user's utterance and presenting a search result according to the request. Is to provide a system.

The present invention adopts the following configuration in order to achieve the above-described object. That is, the present invention provides a search item extraction unit that extracts a recognition word that is a keyword for a search item from voice data including a utterance content of a user who desires an information search,
A voice information extraction unit that extracts emotion information related to a user's emotion from the voice data;
A search control unit that generates and outputs a search expression including a search condition for at least one search item from the recognized word and emotion information;
Is a voice information retrieval system including

  According to the present invention, not only a recognition word as a search keyword but also a search expression including a search condition based on user emotion information is set. Therefore, it is possible to perform a search for grasping a user's request from emotion information and presenting a search result according to the user's request.

Preferably, in the information search system according to the present invention, the search control unit determines weight values for a plurality of search items related to the search based on the emotion information,
A storage unit for storing the weight values;
And a search result evaluation unit that evaluates each search result data using the weighting value stored in the storage unit when at least one search result data matching the search formula is received.

  In this way, when the search result evaluation unit evaluates the search result data using the set weighting and receives a plurality of search result data, the search result evaluation unit responds to the user's request from these. Search results can be extracted and provided.

  Preferably, in the information search system according to the present invention, the voice information extraction unit includes (1) a loudness of a user's voice, (2) a length of silent time included in voice data, and (3) a recognition word. Information indicating at least one of unnecessary words as unrecognized words and (4) the ratio of unknown words contained in the utterance content of the user in the voice data is extracted as the emotion information.

  Further, the present invention can be specified as an information search method, a program, and a computer-readable recording medium on which the program is recorded, having the same characteristics as the information search system described above.

  ADVANTAGE OF THE INVENTION According to this invention, the information search system by the audio | voice which enables the search for grasping | ascertaining a user's request | requirement from a user's utterance content and obtaining the search result according to a request | requirement can be provided.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. The configuration of the embodiment is an exemplification, and the present invention is not limited to the configuration of the embodiment.

[First Embodiment]
<System configuration>
First, a first embodiment of the voice response system according to the present invention will be described. FIG. 1 shows an embodiment of a voice response system according to the present invention. The voice response system shown in FIG. 1 includes, for example, a search item extraction unit 2, a search control unit 3, a search condition storage unit 4, a data storage unit 8, a search result evaluation unit 9, and a message synthesis unit 10. And a search device 14 including the voice information extraction unit 12, and a voice input unit 1 and a message output unit 11 connected to the search device 14.

The voice input unit 1 is configured using a voice input device including a microphone. The voice input unit 1 and the search device 13 may be connected by a network. In this case, as the voice input unit 1, a device having a voice communication (transmission) function such as a personal computer (PC), a workstation (WS), a PDA (Personal Digital Assistants), and a telephone can be applied.

  The message output unit 11 is configured using an audio output device including a speaker. The message output unit 11 and the search device 13 may be connected by a network. In this case, a device having a voice communication (reception) function such as a PC, WS, PDA, and telephone can be applied as the message output unit 11. The message output unit 11 may include a display device that displays information such as a display device in text or an image.

  As described above, the voice input unit 1 and the message output unit 11 can be configured using a PC, WS, PDA, and telephone (a fixed phone and a mobile phone) having a voice input device and a voice output device. The voice input unit 1 and the message output unit 11 may be prepared on the search device 14 side or may be prepared on the user side.

  The voice input unit 1 receives a user's voice for voice search. The voice input unit 1 generates voice data corresponding to the input voice and gives it to the search item extraction unit 2 and the voice information extraction unit 12.

  The search item extraction unit 2 is configured to perform voice recognition processing on the voice data input from the voice input unit 1 and recognize a keyword corresponding to a search item registered in advance as a recognition word. The search item extraction unit 2 gives the search control unit 3 text data of keywords extracted as recognition words. As the search item extraction unit 2, an existing speech recognition engine can be applied.

The voice information extraction unit 12 performs voice recognition processing, sound volume measurement, silence time measurement, and the like on the voice data input from the voice input unit 1, and extracts data related to the user's speech (emotion). For example, the voice information extraction unit 12 is not a portion where the voice in the voice data (content of the user's utterance) is relatively loud, the length of silence (silence time) in the utterance content, and the content requested by the user. User emotion information such as words (unnecessary words for extracting keywords (recognized words): called “unnecessary words (stop words)”), words not registered in the speech recognition system (called “unknown words”), etc. And the extracted emotion information is given to the search control unit 3.

The search condition accumulating unit 4 determines a search condition (keyword) 5 corresponding to a plurality of search items necessary for information search and a search condition relaxation condition (a range to be hit as data that does not completely match the keyword but matches the condition. (Reference) 6 and weighting between search items (standard of which items are emphasized in search) 7 are stored for each user. In addition, the search condition accumulation unit 4 accumulates the voice information reflection reference 13 for each user.

  The search control unit 3 includes a search condition 5, a relaxation condition 6, and a weight 7 stored in the search condition storage unit 4, a recognition word (keyword) input from the search item extraction unit 2, and a voice information extraction unit 12. Based on the obtained emotion information, search method data to be applied to information search is generated and given to the data storage unit 8.

  Specifically, the search control unit 3 sets the keyword itself obtained from the search item extraction unit 2 as it is as a search condition for the search item corresponding to this keyword. In addition, when emotion information is obtained from the voice information extraction unit 12, the search control unit 3 determines relaxation conditions and / or weights according to the content of the emotion information. At this time, the search control unit 3 determines weights such as (1) whether the keyword is obtained by a free dialogue or a dialogue having a scenario, and (2) how many times the keyword is obtained by voice input. It can be considered as an element.

  At this time, if there are already set relaxation conditions and / or weights, these contents (setting values) are changed. In this way, the search control unit 3 generates search conditions and relaxation conditions for each search item finally obtained using the keyword and emotion information as search method data. Then, the search control unit 3 gives the search method data to the data storage unit 8.

  The data storage unit 8 stores search target data. The data storage unit 8 searches for data that matches the search condition and the relaxation condition specified by the search method data given from the search control unit 3, and gives the search result data to the search result evaluation unit 9.

  The search result evaluation unit 9 evaluates the search result obtained from the data storage unit 8 based on the weighting 7 and the voice information reflection reference 13 stored in the search condition storage unit 4, and the search result reflecting the evaluation result The result is output to the message synthesis unit 10. That is, the search result evaluation unit 9 selects a search result to be output from the search result received from the data storage unit 8 based on the weighting 7 and the voice information reflection standard 13, and sends the selected search result to the message synthesis unit 10. Output.

  The message synthesis unit 10 gives various voice messages according to the user's situation to the message output unit. For example, the message synthesis unit gives a guidance (guidance) message related to the use of the voice response system to the message output unit 11. In addition, when the message synthesis unit 10 receives the search result from the search result evaluation unit 9, the message synthesis unit 10 performs a synthesis process with a voice or the like explaining the search result, and sends a message obtained as a result of the synthesis to the message output unit 11. Output.

  The message output unit 11 outputs the voice of the message to the outside. The user can listen to the message output from the message output unit 11. The user can make an answer (voice input) to the voice guidance or inquiry message output from the message output unit 11 using the voice input unit 1. Further, the user can receive the search result from the message output unit 11.

In the search device 14 described above, for example, when the search result evaluation unit 9 determines that further search is necessary based on the evaluation result, a message that prompts input of further search conditions is used instead of the search result. Given to a person. On the other hand, when the user inputs a further search condition by voice, search and evaluation for the search condition are performed. The search result evaluation unit 9 performs a further evaluation result, and gives a search result to the message synthesis unit 10 when a condition for providing the search result to the user is satisfied.

The search device 14 includes a CPU (Central Processing Unit), a main memory (RAM (Random Access Memory), etc.), an auxiliary memory (ROM (Read Only Memory)), an EEPROM (such as a personal computer (PC) and a workstation (WS)). Electrically Erasable Programmable Read-Only Memory), hard disk, etc.) and communication interface, and input devices (keyboard, buttons, pointing device (eg mouse)), and output devices (display, printer, etc.) as necessary Can be configured using a general-purpose or dedicated information processing apparatus (computer).

The functions of each of the search item extraction unit 2, the search control unit 3, the search result evaluation unit 9, and the message synthesis unit 10 and the search function of the data storage unit 8 are, for example, a program stored in the auxiliary memory by the CPU. This is realized by loading it into memory and executing it. Alternatively, each of these units has an ASIC (Application Specific Integrated) for realizing the functions of these units.
It can also be realized by a hardware chip such as Circuit).

  Further, as a storage area of data stored in the data storage unit 8, an auxiliary storage in the search device 14 or a storage area located on a network such as a file server connected to the search device 14 through the network is applied. Can do. The data storage unit 8 can also be configured as an external device that is connected to the search device 14 via a network and returns search results for search method data provided from the search device 14 to the search device.

  Further, as the search condition storage unit 4, a storage device such as a RAM, an EEPROM, or a hard disk can be applied.

  Next, when the voice response system as shown in FIG. 1 is applied to the rental property information search service for real estate, “Rent”, “Room layout”, “Nearest station”, “Time required from the nearest station” ”And“ pet allowed ”will be described as an example in which search conditions are set for five search items and a property is searched.

  FIG. 2 shows an example of dialogue in the rental property information search service. FIG. 2 shows an example in which a user makes a telephone conversation with the search service system. In FIG. 2, when a user calls the system, the system gives a voice question “What kind of property are you looking for?” To the user.

  At this time, the user returns, for example, a reply “for example, one room, (between) about 80,000 yen” to the system. At this time, the system performs a search based on this answer. If the search result obtained at this time does not satisfy the condition for presenting the search result to the user, the system gives the user a question “Is there any other condition?”, For example. In response to this question, the user returns an answer “Please do around Shinjuku Station” to the system.

  Then, the system searches for a corresponding property based on the answer obtained from the user. When the search is completed and the conditions for presenting the search results to the user are satisfied, the system will display a search result saying that the desired property has been **. Return a voice message to the user. Such a dialogue takes place between the system and the user. Hereinafter, information retrieval performed through such dialogue will be described in detail.

  FIG. 3 is a flowchart showing a flow of information search by the voice response system (search device 14) according to the present invention. The flow shown in FIG. 3 starts when the user of the search device 14 accesses the search device 14 using the voice input unit 1 such as a telephone.

  First, the search device 14 makes an inquiry that prompts the user to perform free voice input for the search condition (step S01). For example, the search device 14 gives the user a guidance message for prompting the user to freely input the type and number of items desired to be searched via the message output unit 11. At this time, for example, an inquiry message such as “What kind of property are you looking for?” Shown in FIG. 2 is given to the user.

  Next, the search device 14 receives the voice data related to the answer from the user via the voice input unit 1 (step S02). At this time, for example, the search device 14 receives voice data based on the utterance “One room, (silence) Well, about 80,000 yen” shown in FIG.

Then, the search device 14 uses the search item extraction unit 2 to extract keywords included in the voice data, and the search control unit 3 sets search conditions based on the extracted keywords (step S).
03). In the example of FIG. 2, for example, “One Room” in the answer is set as a search condition for the floor plan, and “80,000 yen” is set as a search condition for the rent.

Further, the search device 14 analyzes the feature of the user's answer (step S04). That is, the voice information extraction unit 12 uses, for example, (1) loudness at the time of answering, (2) timing at which the answer is made, based on voice data related to the answer, that is, the utterance content (speaking method) of the user at the time of answering, And (3)
Unnecessary words, (4) Emotion information such as the ratio of unknown words in answers is extracted.

  Subsequently, the search control unit 3 sets relaxation conditions and weights for the search items based on the extraction result by the voice information extraction unit 12 (step S03). At this time, the search control unit 3 can also apply the inquiry for which an answer has been made (the contents of the inquiry made to obtain the answer) to a change in relaxation conditions or weighting.

  In the example shown in FIG. 2, for example, “about” of “about 80,000 yen” in the answer is extracted as an unnecessary word, and the search condition “80,000 yen” for rent is relaxed (expanding the hit range). And a relaxation condition based on this is set.

  Next, the search device 14 evaluates search target data. That is, the search control unit 3 gives the search condition and the relaxation condition set in steps S03 and S04 to the data storage unit 8, and the data storage unit 8 evaluates the search result using the data that matches the search condition and the relaxation condition as a search result. Part 9 is given. When the search result evaluation unit 9 receives the search result, the search result evaluation unit 9 evaluates the search result using the voice information reflection standard 13 set in the search condition storage unit 4.

Next, the search device 14 performs an information provision / search continuation confirmation process (steps S06, S0).
7). That is, based on the evaluation of the search result by the search result evaluation unit 9, the search device 14 makes a comprehensive determination as to whether to provide the search result information to the user or to continue the search. Such a determination can be made by, for example, the search result evaluation unit 9 or the search control unit 3.

  If information is provided as a result of the determination (S07; Yes), the process proceeds to step S12, the information providing process is performed, and the search result information is transmitted to the user via the message synthesis unit 10 and the message output unit 11. Provided.

  On the other hand, when it is determined that the search is continued (S07; No), the search device 14 analyzes the situation of the user (step S08). For example, it is determined whether or not there is a search item (undecided search condition) for which the search condition is not obtained in the previous answer in other parts in the search control unit 3 or the search device 14 (step S09).

  If there is an undetermined search condition (S09; Yes), the process returns to step S01, and an inquiry for further free answer is made. In the example shown in FIG. 2, an inquiry “Is there any other condition?” Is made. On the other hand, for example, an answer “Please do around Shinjuku” is obtained.

  On the other hand, when there is no undetermined search condition (S09; No), an inquiry about each search item of the search condition is made to the user. When an answer from the user to the inquiry is obtained (step S11), the process proceeds to step S03.

Hereinafter, the process in the flow as described above will be described in detail.
(1) Accumulation of data First, the accumulation of data for performing information retrieval will be described. Search target data to be searched is registered in the data storage unit 8 in advance.

Further, the search condition storage unit 4 of the search device 14 includes (1) the number of articles to be guided as information necessary for the search and its evaluation, and (2) the weight between the items to be evaluated (initial value).
, (3) A function (initial state) that evaluates the search target data for the search conditions of each item, (4) A comprehensive evaluation of the property information using the search conditions entered by the user, weights and functions Functions and (5) threshold values (total evaluation threshold values) that are determined to be close to the user's wishes are registered in advance. These pieces of information (1) to (5) are handled as information constituting the audio information reflection standard 13 used for processing by the search result evaluation unit 9.

  FIG. 4 shows a setting example of the number of cases for guiding a property. FIG. 4 shows a table for registering the number of property guides. In this table, “10” is set as an example of the number of cases. An appropriate value can be set for the number of guidance cases. A table as shown in FIG. 4 can be created on the search condition storage unit 4.

  FIG. 5 shows an example of weighting (initial value) between items to be evaluated. FIG. 5 shows search conditions and relaxation conditions for a plurality of items for evaluation (here, “rent”, “room layout”, “nearest station”, “time required to nearest station”, “pet allowed”). And a table for registering weights.

  In the example illustrated in FIG. 5, the table includes one or more records having items (search items), search conditions, relaxation conditions, and weights as elements. The search conditions, relaxation conditions, and weights registered in this table are handled as search conditions 5, relaxation conditions 6, and weights 7 for the user's search request, respectively. Such a table can be created on the search condition storage unit 4.

  In the example shown in FIG. 5, “3”, “2”, “2” are set as initial values of the weights for the items “rent”, “room layout”, “nearest station”, and “time required to the nearest station”. "," 1 "is set. In the initial state, for example, “not set” is set as the search condition for each item, and “none” is set as the relaxation condition.

  FIG. 6 shows an example of an evaluation function (initial state) for evaluating search target data with respect to the search condition of each item. FIG. 6 shows an example of an evaluation function for the search item “rent”. The parameters of the evaluation function shown in FIG. 6 are an attribute value (here, the amount of rent) corresponding to the search item and the satisfaction level of the user.

Satisfaction with the rent specified by the user at the time of search is used as the reference value. Satisfaction is low if the rent is too low or too high for the designated rent. Specifically, if the rent is lower than the designated rent, the satisfaction level gradually decreases. On the other hand, when the rent is higher than the designated rent, the satisfaction level is drastically reduced. Such an evaluation function is prepared for each item and can be registered in the search condition storage unit 4.

  Four types of patterns (binary type, discrete value type, target type continuous function, non-target type continuous function) corresponding to the item attributes are prepared as functions for evaluating the search target data for each item search condition. Registered in the condition storage unit 4. By these evaluation functions, the user's satisfaction with respect to each item is quantified.

  FIG. 7 shows a pattern of the evaluation function of the search target data. 7A shows a binary evaluation function pattern, FIG. 7B shows a discrete evaluation function pattern, and FIG. 7C shows a symmetric continuous function evaluation function pattern. FIG. 7 (D) shows an evaluation function pattern of an asymmetric continuous function.

As shown in FIG. 7A, the binary type takes one value of “satisfied (condition is satisfied)” and “unsatisfied (condition is not satisfied)”. For example, a search item such as “Can / cannot keep a pet” corresponds to a binary type.

  As shown in FIG. 7B, the discrete value type has a jump value, and the degree of satisfaction increases as the value becomes closer to the value (reference value) specified for the item. For example, a search item such as “nearest station” corresponds to a discrete value type.

  In the symmetric continuous function type, as shown in FIG. 7C, the degree of satisfaction decreases symmetrically as the distance from the value (reference value) specified for the item increases. For example, a search item such as “required time from the nearest station” corresponds to a symmetric continuous function.

  In the non-objective continuous function type, as shown in FIG. 7D, the degree of satisfaction decreases asymmetrically with distance from the value (reference value) specified for the item. For example, a search item such as “rent” corresponds to the asymmetric continuous function type.

  By the evaluation functions of the four patterns as described above, a permissible range (basic permissible range) is defined for each search item even if it slightly deviates from the input search item. These evaluation functions can be created based on statistical values obtained by sample collection such as market research.

  FIG. 8 shows an example of a function that performs comprehensive evaluation on property information using a search condition input by a user, weighting, and an evaluation function. The formula shown in FIG. 8A is a basic comprehensive evaluation formula. In this basic comprehensive evaluation formula, the total score of the products of weighting and satisfaction of each item is treated as a comprehensive evaluation score.

  The formula shown in FIG. 8B is a comprehensive evaluation that places importance on the satisfaction of binary type items. In this comprehensive evaluation formula, the total score of the product of the weight and satisfaction of each item excluding the binary type item is added to the total score of the product of the weight and satisfaction of the binary type item. Are treated as

  As a comprehensive evaluation function (comprehensive evaluation score calculation formula) as shown in FIG. 8, a plurality of calculation formulas having various balances can be prepared in advance in consideration of a balance between expected search items. At this time, the search result evaluation unit 9 can be configured to select an arbitrary calculation formula from a plurality of total evaluation score calculation formulas and use it for the evaluation.

For example, the search result evaluation unit 9 has a search item having the largest weighting value among the set weighting values (if there are a plurality of search items having the largest weighting value, the ranking between items according to a predetermined priority order). Is calculated automatically (for example, the equation shown in FIG. 8B) is automatically selected.

  Alternatively, the search result evaluation unit 9 inputs the search target data using the specified calculation formula according to the calculation formula specification data input from the input device 24 and stored in the search condition storage unit 4 via the input control unit 22. It can also comprise so that comprehensive evaluation of this may be performed.

  FIG. 9 shows an example of threshold values (total evaluation threshold values) that are determined to be close to the user's wishes. FIG. 9 shows a table in which the comprehensive evaluation threshold is registered. The comprehensive evaluation threshold value can be expressed by a numerical value, and a desired value is set as the comprehensive evaluation threshold value.

(2) Extraction of search items Next, extraction of search items will be described. The search item extraction unit 2 of the search device 14 receives audio data corresponding to the audio input to the audio input unit 1. The search item extraction unit 2 extracts a keyword corresponding to the search item as a recognition word by voice recognition processing.

  FIG. 10 is an explanatory diagram of processing by the search item extraction unit 2. As shown in FIG. 10 (A), the user responds with “One room, (between (silence)) Well, about 80,000 yen”, and such voice data is input to the search item extraction unit 2. Assuming that In this case, the search item extraction unit 2 extracts “one room” and “80,000 yen” as keywords (recognized words), and provides the search control unit 3 with data of recognized words.

  Further, for example, when the search item extraction unit 2 receives voice data corresponding to the utterance “Please do so around Shinjuku” by the user as shown in FIG. “Shinjuku” is extracted from the data as a recognized word, and the data of the recognized word is given to the search control unit 3.

(3) Emotion Information Extraction Next, emotion information extraction processing will be described. The voice information extraction unit 12 extracts emotion information (user's speech information) from the voice data of the answer input by the user. FIG. 11 is an explanatory diagram of audio information extraction processing.

The voice information extraction unit 12 is as follows: (1) loudness at the time of answering, (2) timing of answering (silent time), (3) unnecessary words, (4) ratio of unknown words Data that can be acquired from information other than simple recognition words is extracted as emotion information.

  For example, when the voice information extraction unit 12 receives voice data of a user's utterance such as “One room, (silence) Well, about 80,000 yen” as shown in FIG. Information indicating that the “room” and “80,000 yen” were uttered in a relatively loud voice is obtained.

  This is because, in general, when a user makes a reply, important information tends to be uttered in a relatively loud voice so that the other party can fully recognize the important information. In this case, “one room” and “80,000 yen” are important information indicating the desired floor plan and rent, respectively, so they are uttered loudly.

  Also, the voice information extraction unit 12 extracts that there is a silence between “in the one room” and “Em” as information related to the timing of the answer. Also, the voice information extraction unit 12 extracts “Ett” and “about” as unnecessary words.

In addition, when the voice information extraction unit 12 receives voice data of the user's utterance “Please do around Shinjuku” as shown in FIG. 11B, “Shinjuku” is relatively large. Information indicating that the voice is spoken is obtained. In addition, the voice information extraction unit 12 acquires “around” and “please” as unnecessary words.

  At this time, the voice information extraction unit 12 can also acquire the time (silence time) from when the inquiry is made until the answer “around Shinjuku” is obtained as emotion information related to the timing of the answer. . The voice information extraction unit 12 passes the emotion information extracted from the voice data (the content of the user's utterance) to the search control unit 3.

(4) Setting Search Conditions and Relaxation Conditions The recognition words (keywords) extracted by the search item extraction unit 2 are set as search conditions as they are in the search control unit 3. For example, the keyword “one room” extracted in the audio data example shown in FIG. 10 is set as a search condition for the search item “room layout”. The keyword “80,000 yen” is set as a search condition for the search item “rent”. Further, the keyword “Shinjuku” is set as a search condition for the search item “nearest station”.

  On the other hand, the search control unit 3 sets a relaxation condition for the search condition based on the emotion information extracted by the voice information extraction unit 12 (relaxes the search condition). For example, when an unnecessary word indicating the degree or range of the amount of “about” such as “about 80,000 yen” is obtained from the user's answer, the search control unit 3 uses about 80,000 yen For example, 60 to 90,000 yen is set as a relaxation condition for the rent (see the relaxation condition shown in FIG. 13A). Note that unnecessary words corresponding to “about” include “approximately”, “approximately”, “about”, and the like.

In addition, when an unnecessary word indicating a geographical range such as “around” in “around Shinjuku” is obtained from the user's answer, the search control unit 3 requests that the vicinity of Shinjuku station is requested by the user. For example, all the adjacent stations (Okubo Station, Yoyogi Station) on the same line around Shinjuku Station are set as relaxation conditions for the nearest station (relaxation shown in Fig. 13 (B)) See conditions). Note that unnecessary words corresponding to “around” include “near”, “near”, and “about”.

The search control unit 3 determines the content of the relaxation condition according to a predetermined condition. At this time, the search control unit 3 can determine the content of the above-described relaxation condition using an evaluation function corresponding to the search. For example, in the case of rent, in the rent evaluation function obtained when 80,000 yen is set as the reference value, by taking the minimum and maximum values when the satisfaction level is equal to or higher than a predetermined value (for example, 80%), Relaxation conditions such as “60 to 90,000 yen” can be set. In addition, the search control unit 3 uses the evaluation function with Shinjuku Station as a reference value for the nearest station, and the station (
For example, an adjacent station) can be set as the relaxation condition.

(5) Reflection in weighting The search control unit 3 changes the weighting for the search item for evaluating the search result based on the emotion information. In the search condition storage unit 4, as initial values of weighting for each search item, for example, “Rent”, “Room layout”, “Nearest station”, “Time required to the nearest station” as shown in the example of FIG. ”Is preset. Such initial values can be set on the basis of statistical values obtained by collecting samples such as ordinary social conventions and questionnaires.

  When the search control unit 3 receives the emotion information, the search control unit 3 determines whether to reflect the emotion information in the weight based on the emotion information, that is, whether to change the weight value for the search item included in the answer. Judgment is made.

  FIG. 12A shows a determination example in the case where the weighting value for the search item “rent” is reduced based on emotion information, and FIG. 12B shows the weighting value for the search item “room layout” based on emotion information. An example of determination when increasing the value is shown.

  FIG. 12 (A) shows an example in which a reply “Well, about 70,000 yen” was obtained after silence (silence time) after acceptance of the reply was started. In this case, the search control unit 3 performs the following determination, for example.

  Since there is a silent time since the reception of the answer is started, the search control unit 3 determines that the user is worried. In addition, since the unnecessary words such as “Et” and “Waka Kana” are attached before and after the recognition word (keyword) of “70,000 yen”, the search control unit 3 allows the user to set the amount of “70,000 yen”. Judge that it is not strongly hoped. Furthermore, when the volume of the utterance “70,000 yen” is smaller than the average volume of the user or general normal conversation, the search control unit 3 places importance on the amount of money “70,000 yen”. Judge that it is not. The search control unit 3 comprehensively determines these determination results, and determines that the current weighting value for “rent” should be reduced.

  On the other hand, FIG. 12B shows an example in which an answer with a strong tone such as “3LDK!” Is obtained after silence (silence time) after acceptance of the answer is started. In this case, the search control unit 3 performs the following determination, for example.

  Since there is silent time, the search control unit 3 determines that the user is worried. However, since there is no unnecessary word before and after the keyword “3LDK” (stated assertively), the search control unit 3 determines that the user wants “3LDK”. Furthermore, when the utterance “3LDK” is larger than the average volume, the search control unit 3 determines that the user strongly desires “3LDK”. The search control unit 3 comprehensively determines these determination results, and in this case, determines that the current weighting value for the “room layout” should be increased.

  The search control unit 3 changes the weight set in the search condition storage unit 4 based on the determination result of whether or not to change the weight as described above. At this time, the search control unit 3 can consider the number of answers (search items) input and conditions for obtaining the answers (whether or not free search conditions are set).

  FIG. 13A shows an example of weighting reflection in the case where the first response “One room, (silence) Well, about 80,000 yen” is input as shown in FIG. ) Shows a reflection example of weighting in the case where a second response “Please give me around Shinjuku” as shown in FIG. 3 is inputted by voice. However, it is assumed that values as shown in FIG. 5 are set as initial weighting values for each search item. These answers are obtained in a situation where free search conditions can be input.

  As shown in FIG. 13A, when the first response is input, keywords for the search items “room layout” and “rent” such as “one room” and “80,000 yen” are obtained. In this case, the search control unit 3 sets “60 to 90,000 yen” as the relaxation condition of “rent” from the unnecessary word attached to “80,000 yen” of “about”. On the other hand, since “one room” is stated assertively, no relaxation condition is set.

  Subsequently, the search control unit 3 considers that this answer is the first answer for “rent”, and that the keyword that the user particularly emphasizes tends to be input in the first answer. It is determined that 1 is added (+1) to the weighting value for “rent”.

Subsequently, the search control unit 3 considers that the answer “80,000 yen” is obtained under free condition settings, and determines that 1 is added (+1) to the weighting value for “rent”. . On the other hand, the search control unit 3 determines that 1 is subtracted (−1) from the weighting value in consideration of the stagnation of the “rent” rent. Based on these determination results, the search control unit 3 uses the current weighted registration value for “rent” (here, initial value = 3) as (1 + 1−
1) Change to + 3 = 4.

On the other hand, the search control unit 3 determines that “1” is added to the weighting value for “room layout” in consideration of the fact that this answer is the first answer for “room layout”. Further, it is determined that 1 is added to the weighting value in consideration that the search condition “one room” is input in a free voice input environment. Based on these determination results, the search control unit 3 changes the current weighted registration value (here, initial value = 2) to “room layout” to (1 + 1) + 2 = 4.

  In the first response, keywords for the search items “nearest station” and “required time to nearest station” are not input. For this reason, the search control unit 3 does not perform processing for these search items.

  As shown in FIG. 13B, when a second response is input, a keyword for the search item “nearest station” “Shinjuku (station)” is obtained. In this case, the search control unit 3 focuses on Shinjuku Station such as “Okubo Station, Yoyogi Station ...” as a relaxation condition of “Nearest Station”, for example, from an unnecessary word attached to “Shinjuku” called “around”. Set all the stations adjacent to Shinjuku Station on the same line.

  The search control unit 3 does not consider whether or not to add 1 to the weighting based on the assumption that the search condition included in the second and subsequent answers is not emphasized by the user. However, when the search condition included in the second and subsequent answers is obtained under conditions where the search condition can be freely input, 1 is added to the weighting value.

  Accordingly, the search control unit 3 determines that 1 is added to the weighting value in consideration of the fact that the search item “nearest station” included in the second answer is input in a free input environment. In addition, since there is no unnecessary word corresponding to stagnation before “Shinjuku”, there is no negative point for the weighting value. Accordingly, the search control unit 3 changes the current weighted registration value (here, initial value = 2) for “nearest station” to 1 + 2 = 3.

  In the second response, keywords (search conditions) for “rent”, “room layout”, and “time required to the nearest station” are not entered. Therefore, the search control unit 3 does not perform processing for these search items.

(6) Reflection of Relaxation Condition on Evaluation Function When the search item for which the relaxation condition is set based on the voice data of the answer is uttered, the search control unit 3 reflects the relaxation condition on the evaluation function of the search item. As described above, the search condition accumulation unit 4 stores an evaluation function for each search item as one of the voice information reflection criteria 13.

In an initial state (before a user search request is made), a basic evaluation function based on statistical values of sample data is registered. Satisfaction with the value (the value corresponding to the search item) in this basic evaluation function does not match all users.

FIG. 14 is an explanatory diagram showing an example of reflection of the relaxation condition for the evaluation function. FIG. 14 shows an evaluation function for the search item “nearest station”. The basic evaluation function (indicated by a thin line) in this example is a quadratic function that must be a station desired by the user (the user's satisfaction is zero at a station adjacent to the desired station).

On the other hand, it is assumed that the user permits at an adjacent station (when the adjacent station is set as a relaxation condition). In this case, the user's satisfaction with the adjacent station is not zero. Therefore, the search control unit 3 changes the evaluation function so that the user's satisfaction at the station after the desired station (the station next to the adjacent station) becomes zero. Therefore, the evaluation function is a quadratic function as shown by the thick line in FIG.

  When a new relaxation condition is set for the “nearest station” and the evaluation function and the relaxation condition do not match, further correction of the evaluation function (reflection of the relaxation condition) is performed. In this way, each time the relaxation condition is set, the search control unit 3 checks the consistency with the corresponding evaluation function, and at least the satisfaction of the range where the relaxation condition is satisfied does not become zero. Make corrections.

(7) Search When the search control unit 3 sets the search condition and the relaxation condition based on the keyword and emotion information included in the answer, the search control unit 3 relates to the search condition and the relaxation condition, and the AND condition or the OR condition between the search items. A search formula based on data (preset by the user) is created, and the search formula is given to the data storage unit 8 as search method data.

  The search method data includes, for each search item in which a keyword is entered, a relaxation condition if there is a relaxation condition, and a search condition if there is no relaxation condition. These search conditions and / or relaxation conditions are AND conditions or OR conditions, respectively. It is tied with.

  The search control unit 3 can perform a search, that is, assign search method data (search formula) according to a predetermined search start rule. For example, the search control unit 3 can be configured to output search method data each time an answer is obtained. In this case, for example, a search based on the search condition and the relaxation condition shown in FIG. 13A and a search based on the search condition and the relaxation condition shown in FIG. 13B are performed.

  Alternatively, the search control unit 3 can be configured to output the search condition and the relaxation condition when the number of answers reaches a predetermined number as search method data. For example, the search control unit 3 can be configured to output search method data including the search condition and the relaxation condition obtained in the first and second answers shown in FIG.

  Alternatively, the search control unit 3 is configured to generate and output search method data at the time when keywords for all search items for which input is essential are input, or when keywords for all search items are input. be able to. These search start rules can be set as appropriate.

  Upon receiving the search method data (search formula), the data storage unit 8 extracts the search target data satisfying the search formula (search target data hitting the search formula) from the database storing the property information as the search target data. And output as search results. The property information has a record having elements of “rent”, “room layout”, “nearest station”, “time required to nearest station”, and “pet allowed / not allowed”. The search result is given to the search result evaluation unit 9.

(8) Evaluation of Search Property When the search result evaluation unit 9 receives the search result output from the data storage unit 8, the search result storage unit 4 sets the search condition storage unit 4 for individual property data included in the search result. Evaluation is performed using item weights, evaluation functions, comprehensive evaluation functions, and comprehensive evaluation thresholds.

  FIG. 15 shows an example of comprehensive judgment (satisfaction) of property information by the search result evaluation unit 9. FIG. 15A is a table showing relaxation conditions (search conditions), property values, satisfaction levels, and weight values for each search item, which are preconditions for comprehensive judgment. FIG. 15B shows a calculation example of the comprehensive evaluation score, and FIG. 15C shows a determination result as to whether or not the evaluation criterion is satisfied.

  In the table shown in FIG. 15A, the search items are “rent”, “room layout”, “nearest station”, “time required to the nearest station”, and the relaxation conditions (search conditions) are “6˜ They are "90,000 yen", "One room", "Shinjuku", "None", and "None". The weighting values are “3”, “2”, “2”, “1”, and “5”, respectively. In addition, in the table shown in FIG. 15 (A), when the property information having the values of rent “76,000 yen”, floor plan “one room”, and nearest station “Shinjuku” is obtained as one of the search results. Is shown.

  The search result evaluation unit 9 obtains the satisfaction level of the property value corresponding to the relaxation condition using the corresponding evaluation function. In the example shown in FIG. 15A, it is shown that satisfaction [%] of “98”, “100”, and “100” is obtained for each value of the rent, the floor plan, and the nearest station of the property information. Has been. It should be noted that satisfaction level “100” is given to a search item having no relaxation condition.

  Next, the search result evaluation unit 9 calculates a comprehensive evaluation score for the property information by using a comprehensive evaluation function prepared in advance as one of the voice information reflection criteria 13. Here, as the comprehensive evaluation function (computation formula for the comprehensive evaluation score), a formula as shown in FIG. 15B is used for adding the products of the weight values and the satisfaction levels of the respective items. As a result, the search result evaluation unit 9 calculates 1294 points as the overall evaluation points.

  Next, the search result evaluation unit 9 compares the comprehensive evaluation threshold (total evaluation score reference value) prepared in advance with the comprehensive evaluation score, and whether or not the comprehensive evaluation score is equal to or greater than (or exceeds) the comprehensive evaluation threshold. Determine. Here, for example, “1000 points” is registered in advance as the comprehensive evaluation threshold. When the comprehensive evaluation score is equal to or greater than (or exceeds) the overall evaluation threshold, the search result evaluation unit 9 determines that the property information is property information that satisfies the evaluation criteria, and treats the property information as property information to be provided to the user. On the other hand, when the property information does not satisfy the evaluation standard, the property information is discarded without being provided to the user. Since the property information shown in FIG. 15A is larger than the comprehensive evaluation threshold as shown in FIG. 15C, it is treated as property information that satisfies the evaluation criteria.

  In the evaluation, the degree of divergence indicating how far away from the required condition (desired value) can be used as the evaluation value instead of the satisfaction level. In this case, the evaluation function is a function indicating the degree of deviation based on the desired value.

When the search result evaluation unit 9 finishes the evaluation process for all the property information output from the data storage unit 8, the search result evaluation unit 9 or the search control unit 3 provides information and ends the search, Make a comprehensive decision on whether to continue the search. For example, the search result evaluation unit 9 stores property information satisfying the evaluation criteria in advance as the number of property guides registered in advance (for example, 10, FIG. 4).
It is determined that information will be provided.

At this time, the number of property guides is treated as the upper limit of the property information to be provided, and if the number of property information is less than or equal to the number of property guides, all of these property information is provided and property information that satisfies the evaluation criteria If the number of property guides is greater than or equal to the number of property guides, the same number of property information as the number of property guides is provided in descending order of overall evaluation score. However, the number of property guides may be handled as the minimum value of the number of property information to be provided, and all property information that satisfies the evaluation standard more than the number of property guides may be provided.

(9) Presentation of search results Property information that satisfies the evaluation criteria is provided to the user. At this time, the property information may be provided to the user by voice, or may be provided as text or image data and displayed on a display held by the user.

  Search results can be provided in the order of, for example, (1) a property that matches all search conditions, and (2) a property obtained by relaxing the search conditions. At this time, if the number of properties satisfying all the search conditions (not the relaxation conditions) is equal to or greater than the number of property guides, the properties are presented in the order close to the value of the search conditions.

  On the other hand, if the number of properties that satisfy all the search conditions is less than the number of property guides, the properties that match all the search conditions are presented in order from the closest to the search conditions, and then the search conditions are relaxed. From the obtained property, presentation is made in order from the property that matches the item with higher weight. As described above, the search result desired by the user can be obtained by inputting a minimum number of words.

  FIG. 16 shows a search result display example (displays properties that match all search conditions). In FIG. 16, “1LDK of about 100,000 yen” is the search condition, and as the search result when the number of the property guides is 8, eight property information satisfying all the search conditions is “the condition is An example presented as “Property” is shown. Here, the property information is presented in the order close to 100,000 yen as the order close to the search condition.

  FIG. 17 shows a search result display example (displaying a property obtained by relaxing search conditions). In FIG. 17, “1LDK of about 80,000 yen” is a search condition, and as a search result when the number of property guides is 8, three property information satisfying all the search conditions is “the condition is An example is shown in which five pieces of property information that are presented as “property” and then satisfy the relaxation conditions are presented as “next recommended property”.

<Customize search conditions>
Next, a method for defining an evaluation function used for evaluating each search item will be described. An evaluation function used for evaluating each search item stored in the search condition accumulation unit 4 can be defined using a parameter.

18 and 19 are explanatory diagrams of a method for defining an evaluation function for search items. FIG. 18 shows a reference model of an evaluation function for defining the evaluation range of each item. In FIG. 18, in the evaluation function, an attribute value corresponding to an item is set on one axis (x-axis), and a satisfaction level is set on the other axis (y-axis). The reference model sets the desired value of the attribute value to the satisfaction reference value (satisfaction 100
%), It is possible to set how much the set value for the predetermined satisfaction (%) can be set with respect to the desired value.

  In the example shown in FIG. 18, the reference model is a quadratic function that is symmetrical with respect to a straight line parallel to the y-axis when the attribute value has a desired value, and the satisfaction level is 0%, 50%, and 90%. It has an initial setting value. Specifically, the satisfaction level is 90% when the set value of the attribute value is 0.8 times and 1.2 times the desired value. The satisfaction level is 50% when the set value of the attribute value is 0.4 times and 1.6 times the desired value. The satisfaction level is 0% when the set value of the attribute value is 0.0 times and 2.0 times the desired value.

The search device 14 is a reference model of the evaluation function of each item stored in the search condition storage unit 4 (
Display control unit 21 for displaying evaluation function setting screen) on output device (display (display device)) 23
And an evaluation function adjustment unit (input control unit) 22 that receives information input from an input device (keyboard, pointing device (mouse), etc.) 24, performs adjustment processing on the evaluation function, and stores it in the search condition storage unit 4 again. And.

  The administrator (operator) of the search device 14 operates the input device 24 and the output device (display) 23 to call the reference model (evaluation function setting screen) of the evaluation function for each item on the display screen. Can do.

  The evaluation function setting screen has an input form for setting the attribute value for the satisfaction value (0%, 50%, 90%) prepared in advance, and the administrator sets the attribute value (changes the setting value) ), The evaluation function can be changed (adjusted) to a desired shape.

  FIG. 19 shows the setting values (attribute parameters) of the respective attribute values for satisfaction levels 0%, 50% and 90% when the attribute value is larger than the desired value, 1.4 times the desired value, 2.2 The case of changing to double and 4.0 times is shown. Thereby, the shape of the evaluation function is changed as indicated by the broken line.

  A model as shown in FIG. 18 is illustrated as the reference model. On the other hand, the reference model is composed of a straight line having a satisfaction level of 100% regardless of the set value, defines a certain attribute value as a desired value, and has a desired curve by setting the satisfaction level for the attribute value. An evaluation function can also be generated.

<Use examples of unnecessary words>
Next, as another embodiment of the first embodiment, an example in which an unnecessary word is used as a part of emotion information will be described with reference to FIGS.

  FIG. 20 shows a first usage example. As shown in the dialogue example shown in FIG. 20 (A), the system (search device 14) gives an inquiry “What kind of property are you looking for?” Assume that the answer is “approx. 80,000 yen”.

  In this case, processing by the search item extraction unit 2 and the voice information extraction unit 12 is performed on the user's answer, and the search control unit 3 analyzes the utterance (utterance) content of the user. In this analysis, the reply “One room, about 80,000 yen” is divided into words as shown in FIG. As a result, a plurality of words “one room”, “de”, “um”, “80,000 yen”, and “about” are obtained. At this time, “one room” and “80,000 yen” are recognized as recognition words 1 and 2, and “de”, “um” and “about” are recognized as unnecessary words A, B and C, respectively. .

  The search control unit 3 sets search conditions according to the above recognition result. 20C includes a recognition item name (corresponding to the search item in FIG. 5) based on the recognition result in FIG. 20B, a recognition result (corresponding to the search condition in FIG. 5), and a search condition. (Corresponding to the relaxation conditions in FIG. 5).

  The search control unit 3 sets a recognition result and a search condition for the recognition item. For example, as shown in FIG. 20C, the search control unit 3 sets the recognition result (search condition) “80,000 yen” for the recognition item name “rent” based on the recognition word 2 “80,000 yen”. Set. At this time, the search control unit 3 determines that the rent is not definitively specified based on the unnecessary word C “about” uttered after the recognition word 2 (an additional word indicating a range of “about”). So). Then, the upper and lower neighborhood values (for example, 60 to 90,000 yen) of “80,000 yen” are set as the rent search condition (relaxation condition).

Further, the search control unit 3 sets “one room” as the recognition result (search condition) for the recognition item “room layout” from the recognition word 1 “one room”. At this time, the search control unit 3 determines a definitive designation based on the unnecessary word A “de” uttered after the recognized word 1 (the unnecessary word A is
(It ’s not a range word). Accordingly, the search control unit 3 sets “one room” itself as a search condition (relaxation condition) for “room layout”.

  FIG. 21 shows a second usage example. As in the dialog example shown in FIG. 21A, the system (search device 14) gives an inquiry “What kind of property are you looking for?” Assume that the response “within minutes” is returned.

  In this case, processing by the search item extraction unit 2 and the voice information extraction unit 12 is performed on the user's answer, and the search control unit 3 analyzes the utterance (utterance) content of the user. In this analysis, the answer “Eh, within 5 minutes from Shinjuku station” is divided into words as shown in FIG. As a result, a plurality of words “e”, “Shinjuku”, “around”, “from the station”, and “within 5 minutes” are obtained. At this time, “Shinjuku” and “within 5 minutes” are recognized as recognition words 1 and 2, and “Eh,” “around,” and “from station” are recognized as unnecessary words A, B and C, respectively. The

In addition, “within” is a word indicating a range, but is a word indicating an upper limit of an attribute value, and thus is extracted and recognized as a recognized word. In addition, a word indicating a lower limit such as “above” is extracted and recognized as a recognized word together with a word as an attribute value. Such recognition depends on the registered contents of the recognition word (keyword).

  The search control unit 3 sets search conditions according to the above recognition result. The table shown in FIG. 21C includes a recognition item name (corresponding to the search item in FIG. 5) based on the recognition result in FIG. 21B, a recognition result (corresponding to the search condition in FIG. 5), and a search condition. (Corresponding to the relaxation conditions in FIG. 5).

  The search control unit 3 sets a recognition result and a search condition for the recognition item. For example, as shown in FIG. 21C, the search control unit 3 sets a recognition result (search condition) “Shinjuku” for the recognition item name “closest station” based on the recognition word 1 “Shinjuku”. At this time, the search control unit 3 determines that it is not an affirmative designation of the nearest station based on the unnecessary word B “around” uttered after the recognized word 1 (an attached word indicating a range of “around”). So). In this case, the search control unit 3 employs the adjacent station of Shinjuku Station as a search condition (relaxation condition) for “nearest station”.

  The search control unit 3 sets “within 5 minutes” as the recognition result (search condition) for the recognition item “required time to the nearest station” from the recognition word 2 “within 5 minutes”. At this time, the search control unit 3 determines that the designation is definitive from the utterance of the recognized word 2 (there is no unnecessary word after the recognized word 2, etc.). Accordingly, the search control unit 3 sets “within 5 minutes” itself as a search condition (relaxation condition) for “time required to the nearest station”.

  FIG. 22 is a table showing an example when search conditions are set based on unnecessary words. FIG. 22 shows examples of unnecessary words added to the recognition words (keywords) for the search items “rent”, “room layout”, “nearest station”, and “time required to the nearest station”, and the unnecessary words. A table showing an example of setting search conditions (relaxation conditions) when is added is shown.

<Weighting rules based on unknown words (unnecessary words) and silence (silence)>
Next, weighting rules based on unknown words and silence (silence time) will be described. FIG.
A) is a table showing an example of a weighting rule based on unknown words, FIG. 23B is a table showing an example of a weighting rule based on silence, and FIG. 23C is an example of weighting based on silence. It is explanatory drawing shown.

  FIG. 23A shows conditions (scenes and detailed conditions), weight adjustment targets, weight change values, and reasons as weighting rules for unknown words. For example, whether the answer is the first inquiry or the second or later is classified as a condition, and further, as a detailed condition, the ratio of unknown words is 70% or more or 50% of all recognized words (the content of utterances subjected to speech recognition processing). It is divided into the above. Furthermore, desired conditions (items corresponding to keywords) recognized from the same utterance as the unknown word are set as adjustment targets, and negative values are respectively set as weighting values. The negative value is because there is a high possibility that the user uttered while hesitating.

  In the example shown in FIG. 23B, when two or more search conditions (keywords) are included in the user utterance (answer), (1) silence period (silence) of 2 seconds or more between the search conditions. When there is, (2) weighting is adjusted on condition that there is a silence period longer than 1 second and shorter than 2 seconds between search conditions.

  In the above-described adjustment conditions (1) and (2), the search condition (search item) immediately after the silent period is a weighting adjustment target. For the adjustment conditions (1) and (2), the weight adjustment values “−2” and “−1” are prepared because the user may have lost the utterance of the search condition. ing.

  For example, as shown in FIG. 23C, recognition word 1, unnecessary word A, recognition word 2, and unnecessary word B are recognized from the answer, and silent time is detected between the recognition word and unnecessary word A. In such a case, the search control unit 3 satisfies the adjustment time (1), satisfies the adjustment condition (2), does not satisfy the adjustment conditions (1) and (2) (the silence time is less than 1 second). If one of the adjustment conditions (1) and (2) is satisfied, a prescribed weighting adjustment is performed for the search condition corresponding to the recognition word 2.

Furthermore, as shown in FIG. 23B, for example, (3) weight adjustment is performed on the condition that there is a silence period of 2 seconds or more before the answer is uttered. In adjustment condition (3), user utterance (
All search conditions (search items) included in (Answer) are subject to weight adjustment. The weight adjustment value “−1” is prepared for the reason that it is possible that the utterance has started without the desired condition being clearly determined and the determination that the desired search condition is not assertive. When the adjustment condition (3) is satisfied, the search control unit 3 performs weighting adjustment (−1) for all the search conditions included in the answer.

<Operation and Effect of First Embodiment>
As described above, according to the search device 14, not only the keyword corresponding to the search item is extracted by the search item extraction unit 2, but also the emotion information related to the user's way of speaking is extracted by the voice information extraction unit 12. And the search control part 3 sets the relaxation condition of search conditions based on emotion information. As a result, the search can be performed under the relaxed conditions reflecting the user's intention. Therefore, the user does not need to be particularly aware of the input of relaxation conditions.

  Further, according to the search device 14, the search control unit 3 determines (changes) the weighting for each item based on the emotion information (eg, stagnation and silence time). The weight value for each search item is used by the search result evaluation unit 9 for evaluation of each search data (property information) searched from the data storage unit 8 together with the evaluation function. Thus, information reflecting the user's intention obtained from the user's emotion information can be provided to the user.

In addition, when the answer is the first answer, the search control unit 3 adjusts the weighting value for the search condition included in this answer. Furthermore, when the answer input is performed in a free voice input environment, the search control unit 3 adjusts the weighting for the search condition input in such an environment. The weighting value is used for evaluation of search data as described above. As a result, information reflecting the answer status of the user can be provided to the user.

  In addition, since the search device 14 has an evaluation function creation support function and provides an evaluation function setting screen to the administrator, the administrator of the search device 14 sets, registers, or changes an evaluation function for each item. The burden in the event of a loss is reduced.

[Second Embodiment]
In the first embodiment, the retrieved data is evaluated using the generally-acquired speech information evaluation standard 13. On the other hand, using the voice information stored for each individual in the search device 14, it is possible to grasp the emotions of individual users in detail and provide feedback to the search. 2nd Embodiment demonstrates the example which evaluates a search result using the audio | voice information for every individual.

  FIG. 24 is a diagram illustrating a configuration example of a voice response system (information retrieval system) according to the second embodiment. In FIG. 24, the same components as those in the first embodiment shown in FIG. FIG. 25 is a diagram illustrating a setting example of weighting using personal attribute information (personal information).

  In FIG. 24, the search device 14 </ b> A according to the second embodiment further includes a personal attribute storage unit 15. The personal attribute storage unit 15 is connected to the voice information extraction unit 12 and the search control unit 3 and stores personal attribute information based on emotion information extracted from the voice information extraction unit 12. The personal attribute information stored in the personal attribute storage unit 15 is used as the weighting 7 stored in the search condition storage unit 4 and used as one of the audio information reflection standards 13. The personal attribute storage unit 15 is created on a storage medium included in the search device 14.

In the second embodiment, data storage in the search condition storage unit 4 and search condition extraction by the search item extraction unit 2 are performed in the same manner as in the first embodiment. On the other hand, in the voice information extraction process performed by the voice information extraction unit 12, when the voice information extraction unit 12 receives the voice data input from the voice information input unit 1, the voice print pattern of the voice data is stored as a personal attribute. The voice print pattern registered in the unit 15 is collated to identify the voice data utterer (individual, ie, user).

The personal voiceprint pattern is created from, for example, voice data obtained when the individual uses the information search system for the first time, and is registered in the personal information storage unit 15 in association with the personal information of the individual. Note that a method of using a caller number (a method of specifying a user from a caller number of a user (for example, a telephone number)) can also be applied to individual identification. After identifying the individual, the voice information extraction unit 12 performs the same processing as that described in the first embodiment.

  Similar to the first embodiment, the search control unit 3 changes the weighting value of the search item based on the emotion information extraction result from the audio data. However, in the second embodiment, the weighting information (personal information) prepared for each individual in the personal attribute storage unit 15 is not the initial value of the weighting value generally prepared as described in the first embodiment. Is used to determine (change) the weighting value.

FIG. 25 shows a setting example of weight values prepared for individuals. In the personal information storage unit 15, initial values of personal information as shown in FIG. 25 (1) are registered. In the initial state, for example, “3” is set as a general weighting value for each search item of “rent”, “room layout”, “closest station”, “time required to the nearest station”, and “pet allowed / impossible”. “,” “2,” “2,” “1,” and “5” are set respectively (FIG. 25 (1)).

Thereafter, when the individual uses the first search using the search device 14 </ b> A, the initial value of the personal information is read from the personal information storage unit 15. The search control unit 3 changes the personal information (weighted value of each search item) read from the personal information storage unit 15 based on the contents of the answer (emotion information, answer status, etc.). The changing method itself is the same as in the first embodiment. For example, as shown in FIG. 25 (2), the weight values of the respective retrieval items read from the personal information storage unit 15 are “5”, “4”, “2”, “1”, “5”. Be changed.

  Thereafter, information search is performed based on the search condition obtained from the answer, and the evaluation of the search result is performed in the same manner as in the first embodiment based on the changed weighting value. Thereafter, for each search item, for example, an average value of the first weighting value (the weighting value before the change) and the weighting value after the change is calculated, and the average value of each search item is changed to a new weighting value (personal information ) Is registered in the personal information storage unit 15 (see FIG. 23 (3)). Note that a calculation method different from the calculation of the average value can be applied as a new weighting value calculation method.

  Thereafter, in the second and subsequent use of the individual, the weighting value is changed based on the emotion information included in the answer and the answer status by the same method as in the first use (FIG. 25 (4)), and the search ends. Later, the average value of the weight values of each search item is registered in the personal information storage unit 15 as a new personal weight value (FIG. 25 (5)).

  According to the second embodiment, the first embodiment uses a generally prepared value as a weighting value to be changed, whereas a weighting value (personal information) prepared for each individual is to be changed. The weight value is changed (adjusted) as follows. That is, an adjustment is made to the weighting value in which the individual attribute (emotion) is reflected in advance. Therefore, compared with the first embodiment, the intention of the individual can be strongly reflected in the evaluation of the search result, and the individual can receive an appropriate search result (information provision).

[Third Embodiment]
Next, as a third embodiment, an example in which audio information acquired by the audio information extraction unit 12 is fed back to navigation provided at the time of search as well as reflected in the search will be described. FIG. 26 is a system configuration diagram of the third embodiment, showing an example in which audio information is applied to both result evaluation and navigation. Note that in the third embodiment, identical symbols are assigned to configurations similar to those in the first embodiment and descriptions thereof are omitted.

  26, the search device 14B further includes a scenario data storage unit 16 and a navigation processing unit 17 in addition to the configuration described in the first embodiment (see FIG. 1). The navigation processing unit 17 performs information retrieval navigation according to the scenario data stored in the scenario data storage unit 16.

  That is, the navigation processing unit 17 includes an input state analysis unit 18 and a scenario analysis unit 19. The input state analysis unit 18 receives the emotion information extracted by the voice information extraction unit 12 and obtains a certainty factor for the user's answer. The scenario analysis unit 19 selects an appropriate piece of navigation message information registered in advance as scenario data in the scenario data storage unit 16 and gives it to the message synthesis unit 10. As a result, a navigation message corresponding to the certainty factor of the user can be given to the user, and it can be determined whether or not to perform a search with the current search conditions.

FIGS. 27A to 27C show examples of navigation message information registered in the scenario data storage unit 16 when asking the user for additional search items. FIG. 27A shows a message “Now I will guide the search results as is” as a navigation message (navigation 1) when the certainty level indicates “strong confidence”.

  FIG. 27B shows a message “Is there other conditions?” As a navigation message (navigation 2) when the certainty level indicates “ordinary certainty”. FIG. 27C shows a navigation message (navigation 3) when the certainty level indicates “weak belief” in the range of “XX to YY (“ XX ”and“ YY ”indicate arbitrary values). "Let's look for it" message.

  FIG. 28 shows an example of a method of calculating the user's certainty from voice information (emotion information) (quantification of certainty). As shown in FIG. 28 (A), the voice information extraction unit 12 is configured to obtain voice volume, silent time, unnecessary words, and the like from voice data input from the voice input unit 1 as in the first embodiment. Emotion information is extracted and provided to the input state analysis unit 18.

As shown in FIG. 28B, the input state analysis unit 18 sets the initial value of the certainty factor (in this example, “
When the emotion information is input, the input state analysis unit 18 increases or decreases the certainty factor based on the user's voice information as shown in FIGS. 29 (A) and (B), for example. According to the rules of
Increase / decrease values (defined in advance) for certainty levels for loudness, silent time, and unnecessary words. The input state analysis unit 18 has information related to such rules in advance.

  FIG. 29A shows an example of a rule for increasing or decreasing the certainty factor based on the loudness of the voice. In this example, the detected sound is compared with a predetermined reference sound, and if the ratio of the volume of the sound to the reference sound is equal to or less than a predetermined value (certainty reduction threshold), the certainty factor decreases and the predetermined value (the certainty factor) If it is equal to or higher than the (rising threshold), the certainty factor is increased.

  In the example shown in FIG. 29 (A), for example, when the sound of the detected recognition word is compared with the reference sound and the volume of the sound is 60% or less of the reference sound, The value decreases by the confidence adjustment value. For example, the certainty factor is “−2”. This is because the user may be lost.

  On the other hand, when the volume of the sound is 120% or more of the reference sound, the certainty value increases by the certainty adjustment value. For example, the certainty factor is “+2”. This is because the user seems to be confident.

  FIG. 29B shows an example of a rule for increasing or decreasing the certainty factor based on the silent time. In this example, it was detected that there was a predetermined length of silence between the second and subsequent conditions, and that there was a predetermined length of silence before the entire input speech. When detected, the certainty factor is configured to decrease.

In the example shown in FIG. 29B, when a silent time of 1 second or longer is detected between conditions, the certainty factor decreases ("-1" from the current certainty factor value). This is because the user may be lost. In addition, when it is detected that the silent period has been 2 seconds or more before the voice is input after the reception of the voice input is started, the certainty factor decreases (1 from the current certainty factor value). Reduced). This is because the desired condition may have started without being clearly determined, and it can be determined that the desired condition is not assertive.

Returning to FIG. 28B, the input state analysis unit 18 subtracts 2 from the current certainty factor, for example, when the loudness of the recognized word is small. The input state analysis unit 18 adds 1 to the current certainty factor, for example, when the answer timing is the earliest (when the recognized word is uttered immediately after the silent time). Furthermore, the input state analysis unit 18 subtracts the number of unnecessary words from the current certainty factor when there is an unnecessary word (for example, “Et”, “Soda”, “Um”) that indicates hesitation.

Thereby, the input state analysis unit 18, for example, as shown in FIG. 28 (B), elements of emotion information specified in advance (for example, loudness, silent time (answer timing), and unnecessary words ), Increase / decrease values “−2”, “+1”, and “−1” are obtained, respectively. Then, the input state analysis unit 18 changes (adjusts) the initial value of the certainty factor (current certainty factor value) “5” to “3”. The adjusted certainty factor is given from the input state analysis unit 18 to the scenario analysis unit 19.

  The scenario analysis unit (corresponding to the confidence level acquisition unit and the navigation message selection unit) 19 determines the degree of confidence (level) based on the confidence value provided from the input state analysis unit 18. For example, the scenario analysis unit 19 has a threshold value (reference value) at a level corresponding to the certainty value. For example, when three levels of “strong”, “normal”, and “weak” are prepared, (number of levels−1) reference values for obtaining a level for the certainty factor are prepared.

  As shown in FIG. 27, when three levels of “strong belief”, “ordinary belief”, and “weak belief” are prepared, for example, “8” and “ 3 "is prepared. Then, the scenario analysis unit 19 compares the obtained certainty factor (“3” in the example of FIG. 28) with each reference value to obtain a certainty level.

  FIG. 30 shows an example of a criterion for the scenario analysis unit 19 to obtain the confidence level. In the example illustrated in FIG. 30, if the certainty value is 8 or more, it is determined that the certainty is strong. If the certainty value is 3 or less, it is determined that the certainty is weak. If the certainty value is greater than 3 and less than 8, it is determined that the certainty is normal.

  When the scenario analysis unit 19 obtains the level of confidence, the scenario analysis unit 19 reads a message corresponding to the level of confidence from the navigation messages stored in the scenario data storage unit 16 as illustrated in FIG. To give. As a result, a navigation message corresponding to the level of confidence is provided to the user in the dialog between the user and the system (search device 14B).

  According to the third embodiment, the user's confidence level is obtained from the emotion information obtained from the user's answer, and a navigation message corresponding to this level is provided to the user. For example, when it is recognized that the user has strong confidence in the answer, the input of the search condition can be terminated at that time. This makes it possible to provide information retrieval in a short time.

In addition, if the level of confidence is recognized as normal or weak, by searching for other search conditions or making a search suggestion, the user's intention (Emotion) can be reflected. Therefore, accurate search results can be provided to the user in a short time.

  The configurations in the first to third embodiments described above can be appropriately combined without departing from the object of the present invention.

[Effect of the embodiment]
According to the first to third embodiments described above, the following effects can be expected in the field of voice search. (A) Accurate search results can be presented without visually confirming search items or setting many search items at once. (B) Even if the user is not conscious of setting fine search conditions such as weighting, range setting, and 1 item 2 conditions, the system (search device) automatically sets and presents the search results. (C) If the user performs a minimum setting (speech input) such as reading out setting items, the result necessary for him can be obtained in a short time. (D) Appropriate search results can be obtained using the evaluation function, without being limited to the priority of search items and the selection of relaxation items.

  Furthermore, the following effects can be obtained as a secondary effect. (E) By performing an accurate search, access to a search service having this system increases. (F) By presenting accurate recommended information, it is possible to lead to an increase in sales of products as search results.

[Others]
The above-described embodiment of the present invention discloses the following invention. In addition, the invention shown below can be combined suitably as needed.

(Supplementary Note 1) A search item extraction unit that extracts a recognition word as a keyword for a search item from voice data of a user who desires an information search;
A voice information extraction unit that extracts emotion information related to a user's emotion from the voice data;
A search control unit that generates and outputs a search expression including a search condition for at least one search item from the recognized word and emotion information;
Information retrieval system using voice. (1)

(Additional remark 2) The said search control part determines the weighting value with respect to the some search item which concerns on search based on the said emotion information,
A storage unit for storing the weight values;
Appendix 1 further includes a search result evaluation unit that evaluates each search result data using the weighting value stored in the storage unit when at least one search result data matching the search formula is received. Information retrieval system by voice. (2)

(Additional remark 3) The said search control part is an information search system by the voice of Additional remark 2 which determines the weighting value with respect to these search items further based on the condition when the said audio | voice data were obtained.

(Supplementary Note 4) The voice information extraction unit includes (1) loudness of the user's voice, (2) length of silent time included in the voice data, and (3) an unnecessary word as a word that is not recognized as the recognized word. , (4) information indicating at least one of the ratios of unknown words included in the utterance content of the user in the voice data is extracted as the emotion information.
The information search system by voice according to any one of appendices 1 to 3. (3)

(Supplementary Note 5) When the voice information extraction unit extracts a voice level for a keyword of a user whose volume is smaller than a predetermined value, the search control unit weights the search item corresponding to the keyword. The information search system by voice according to the supplementary note 4, which reduces the value.

(Additional remark 6) When it is recognized that utterance with respect to the said keyword is performed decisively based on the said emotion information, the said search control part increases the weighting value of the search item corresponding to this keyword. ,
The information search system by voice according to appendix 4 or 5.

(Supplementary Note 7) When the voice information extraction unit detects a silent time longer than a predetermined time between two keywords included in the voice data, the search control unit searches for a keyword corresponding to the rear side. The information search system by voice according to any one of appendices 4 to 6, wherein the weighting value of the item is reduced.

(Additional remark 8) When the silence information longer than predetermined time is detected before the first utterance included in the said audio | voice data in the said audio | voice information extraction part, the said search control part is all the audio | voice data included in this audio | voice data Decrease the weight of each search item corresponding to the keyword,
The information search system by voice according to any one of appendices 4 to 7.

(Additional remark 9) When the said search control part extracts the unnecessary word which shows the range or grade with respect to the attribute value of the search item specified by the said keyword in the said audio | voice information extraction part, the attribute specified by this keyword Create search criteria with relaxed values,
The information search system by voice according to any one of appendices 4 to 8.

(Additional remark 10) The said search control part is a search corresponding to this keyword, when the unnecessary word which shows that the user is troubled in the said voice information extraction part is extracted before and / or after the said keyword. Decrease the item weight value,
The information search system by voice according to any one of appendices 4 to 9.

(Additional remark 11) The said search control part respond | corresponds to the keyword contained in this utterance content, when the ratio of the unknown word contained in the user's utterance content extracted by the said audio | voice information extraction part exceeds a predetermined value. The information search system by voice according to any one of appendices 4 to 10, wherein the weight of the search item is reduced.

(Supplementary Note 12) The storage unit further stores an evaluation function for each search item,
The search information evaluation unit according to any one of supplementary notes 2 to 11, wherein the search result evaluation unit evaluates the search data based on the weighting value and the evaluation function.

(Additional remark 13) The said search result evaluation part is an information search system by voice of Additional remark 12 which performs the said evaluation using the evaluation function which has the evaluation function pattern according to the attribute of the search item.

(Additional remark 14) The said search result evaluation part uses the evaluation function according to a search item for every search item contained in a search formula, The user satisfaction with respect to the attribute value of the search item which search result data have, or The degree of divergence from the attribute value desired by the user is obtained, and an overall evaluation score for the search result data is obtained using the degree of satisfaction or divergence for each of the obtained search items and the weighting value. The information search system by voice according to appendix 12, wherein search result data that is equal to or greater than the threshold value or exceeds the predetermined threshold value is extracted as search result data to be provided to the user.

(Supplementary Note 15) A display control unit that displays the evaluation function stored in the storage unit on a display device;
An evaluation function adjustment unit that accepts an attribute value of a search item for the satisfaction or divergence input using the evaluation function displayed on the display device, adjusts the evaluation function, and stores the evaluation function in the storage unit;
The information search system by voice according to appendix 13, further comprising:

(Additional remark 16) The said search result evaluation part uses the calculation formula which satisfy | fills predetermined conditions from the calculation formulas of the said total evaluation score in which the balance of evaluation between several search items was considered. The information search system by voice according to Supplementary Note 14 or 15 for calculating points.

(Additional remark 17) The personal information storage part which stores the weighting value with respect to each search item prepared for every user is further included,
When the search control unit determines the weighting value of the search item based on the emotion information, any one of the appendices 2 to 16 that increase or decrease the weighting value corresponding to the search user stored in the personal information storage unit Information retrieval system by voice as described in Crab.

(Supplementary note 18) The voice information according to supplementary note 17, further comprising means for updating the weight value of the user stored in the personal information storage unit with a weight value reflecting the weight value increased or decreased by the search control unit. Search system.

(Supplementary note 19) Based on the emotion information extracted by the voice information extraction unit, a belief level acquisition unit for obtaining a belief level with respect to the utterance content of the user;
The information search by voice according to any one of appendices 1 to 18, further including a navigation message selection unit that selects a navigation message related to the search prepared according to the level of confidence and outputs the navigation message to provide to the user system.

(Supplementary note 20) The voice information search system according to supplementary note 19, wherein the navigation message selection unit selects and outputs a navigation message indicating that the input of the search condition is terminated according to the level of confidence.

(Supplementary note 21) The voice information search system according to supplementary note 19 or 20, wherein the navigation message selection unit selects and outputs a navigation message that prompts input of another search condition according to the level of certainty.

(Additional remark 22) The said navigation message selection part selects and outputs the navigation message for confirming the search in a predetermined range about a predetermined | prescribed search item according to the said confidence level, Additional remarks 19-21 Information retrieval system by voice.

(Supplementary note 23)
From the voice data of users who want to search for information, we extract the recognition words that are keywords for the search items,
Extracting emotion information related to the user's emotion from the voice data,
Generating and outputting a search expression including a search condition for at least one search item from the recognition word and emotion information;
Information retrieval method using voice. (4)

(Supplementary Note 24)
Extracting a recognition word as a keyword for a search item from voice data of a user who desires information search;
Extracting emotion information relating to the user's emotion from the voice data;
Generating and outputting a search expression including a search condition for at least one search item from the recognition word and emotion information;
Or a computer-readable recording medium on which the program is recorded. (5)

1 is a system configuration diagram of a first embodiment of an information search system according to the present invention. The figure which shows the example of a user's interaction with the system performed in the information retrieval service realized by the information retrieval system Flow chart showing the flow of information search in the information search system A figure showing an example of setting the number of cases to guide a property Table showing examples of weights (initial values) between search items for evaluating search result data The figure which shows the example of the evaluation function (initial state) used for evaluation of search result data Explanatory diagram of evaluation function pattern Diagram showing an example of a function for comprehensive evaluation (comprehensive evaluation calculation formula) The figure which shows the example of comprehensive evaluation threshold Diagram showing examples of data that can be acquired from keywords (recognized words) Diagram showing examples of data (emotion information) that can be acquired from information other than keywords (recognition words) The figure which shows the example of judgment of whether to reflect to "weighting" based on emotion information Diagram showing an example of reflection to weight Diagram showing an example of how relaxation conditions are reflected in the evaluation function Figure showing example of comprehensive judgment (satisfaction) of property information Diagram showing search result presentation example (showing properties that match all search conditions) Figure showing search result presentation example (showing properties that match all search conditions, and then presenting properties obtained by relaxing the search conditions) Diagram showing an example of setting the evaluation range for each item The figure which shows the example of setting the attribute parameter of the evaluation function Explanatory drawing of an example (use example 1) of using unnecessary words as part of emotion information Explanatory drawing of an example (use example 2) of using unnecessary words as part of emotion information Table showing an example of determining search conditions (relaxation conditions) based on unnecessary words The figure which shows the example of the rule which sets weighting based on an unknown word or silence time The figure which shows the system configuration example of 2nd Embodiment (performing evaluation using personal information) of the information search system by this invention The figure which shows the example of the setting of the weight using personal information The figure which shows the system configuration example of 3rd Embodiment (navigation change by the content of voice input) of the information search system by this invention Figure showing an example of registered navigation messages A figure showing an example of quantification of the user's confidence (a figure showing how to obtain confidence) The figure which shows the example of the rule of certainty increase / decrease based on user's voice information (emotion information) The figure which shows the judgment example of the level of belief based on the belief degree

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Voice input part 2 Search item extraction part 3 Search control part 4 Search condition storage part (storage part)
8 Data storage unit 9 Search result evaluation unit 10 Message composition unit 11 Message output unit 14 Search device (information search system)
15 Personal attribute storage (Personal information storage)
16 Scenario data storage unit 17 Navigation processing unit 18 Input state analysis unit 19 Scenario analysis unit (confidence level acquisition unit, navigation message selection unit)
21 Display Control Unit 22 Evaluation Function Adjustment Unit (Input Control Unit)
23 Display (display device)
24 input devices

Claims (5)

A search item extraction unit for extracting a recognition word as a keyword for a search item from voice data of a user who desires information search;
A voice information extraction unit that extracts emotion information related to a user's emotion from the voice data;
A search control unit that generates and outputs a search expression including a search condition for at least one search item from the recognized word and emotion information;
Information retrieval system using voice. The search control unit determines weight values for a plurality of search items related to the search based on the emotion information,
A storage unit for storing the weight values;
The search result evaluation part which evaluates with respect to each search result data using the weighting value stored in the said storage part when the at least 1 search result data matching the said search expression is received. Information retrieval system using written voice. The voice information extracting unit includes (1) loudness of the user's voice, (2) the length of silent time included in the voice data, (3) an unnecessary word as a word that is not recognized as the recognized word, (4) Information indicating at least one of the ratios of unknown words included in the utterance content of the user in the voice data is extracted as the emotion information.
The information search system by voice according to claim 1 or 2. Computer
From the voice data of users who want to search for information, we extract the recognition words that are keywords for the search items,
Extracting emotion information related to the user's emotion from the voice data,
Generating and outputting a search expression including a search condition for at least one search item from the recognition word and emotion information;
Information retrieval method using voice. On the computer,
Extracting a recognition word as a keyword for a search item from voice data of a user who desires information search;
Extracting emotion information relating to the user's emotion from the voice data;
Generating and outputting a search expression including a search condition for at least one search item from the recognition word and emotion information;
A program that executes

Images