JPH04332084A - Automatic information offering method - Google Patents

Abstract

PURPOSE:To automatically execute the selection offer of proper information corresponding to a subject by identifying the fields of subjects at that time, explicit subjects and relational subjects when an interaction or presentation executor acquires information. CONSTITUTION:A speech based upon interaction or presentation is converted into a character string by a microphone 21 and a speech recognizing device 22, the character string is divided in each word as word strings by a morpheme analyzing device 23 to identify the part of speech of each word, the conjugation of a conjugation word, and so on. Surface language expression such as a mode of expression and a subject converting position based upon information exchange between hierarchies are detected by a field identification/field conversion detector 24, an explicit subject conversion detector 25, an explicit subject identifying device 26, and a relational subject identifying device 27 and a subject in a segment continuing the same subject in respective heirarchy is identified. Consequently the selection and offer of proper information corresponding to the contents of a subject can be automatically executed.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the provision of information to persons conducting dialogue and presentations.

0002

[Background Art] Conventionally, in order to promote dialogues and presentations, those who carry out such activities either predict the topic in advance and select relevant information, or search for information when the information is needed. It is necessary to provide necessary information during dialogues and presentations.

0003

[Problems to be Solved by the Invention] However, since it is generally very difficult to predict the content of a dialogue or presentation in advance, it is difficult to prepare all relevant information in advance. Moreover, when performing a search during a conversation or presentation, the search is performed from a huge database containing a variety of information, so it takes time to obtain the necessary information and the flow of the discussion is interrupted. There is a problem with putting it away.

[0004] In addition, when a device is used to identify the topic of a dialogue or presentation and use it to provide related information, it is also possible to The problem was that the accuracy of topic identification was poor, and appropriate information could not be provided because the three layers of topic identification and related topics developed in relation to that topic were mixed up.

[0005] The present invention has been made to solve the above-mentioned problems, and an object of the present invention is to improve the passive or active information acquisition behavior of the person performing the dialogue or presentation at that time. Automatically select and provide appropriate information according to the topic by identifying the topic field, a newly introduced explicit topic, and related topics that are developed in relation to that topic. Our goal is to provide technology that enables

The above and other objects and novel features of the present invention will be made clear by the present specification and the accompanying drawings.

[0007]

[Means for Solving the Problems] In order to achieve the above-mentioned purpose, in an automatic information providing method that performs morphological analysis of natural language input from an input device and searches an information database based on the analysis result, Using the field terminology dictionary, we identify topics in each of the three layers of the field, explicit topic, and topics related to the explicit topic, determine the severity of the identified topics, and select those with high likelihood. The main feature is that the database is searched for from the database.

[0008] Furthermore, information regarding topic changes detected in the three layers is mutually communicated.

[0009]

[Operation] According to the above-mentioned means, for each of the three layers: topic field, explicit topic newly introduced, and related topics developed in relation to that topic, superficial expressions such as phrases, etc. Automatically selects and provides appropriate information according to the content of the conversation by detecting topic change points using linguistic expressions and information exchange between layers, and identifying topics in segments where the same topic continues in each layer. This allows for effective promotion of dialogue and presentation progress. That is, since information related to the topic of the person conducting the dialogue or presentation is made available for search in a short time, information can be provided in a short period of time during the dialogue, and the topic further develops. Of course, it is also possible to provide information about the speaker's topic in real time.

0010

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

FIG. 1 is a flowchart showing the overall process flow of an embodiment of the automatic information providing method of the present invention, and FIG.
FIG. 1 is a block diagram showing a schematic configuration of an automatic information providing system for implementing the automatic information providing method of the present embodiment.

First of all, the overall flow of information provision processing will be explained with reference to FIG. Each number in FIG. 1 will be called a step.

First, in step 1, utterances in dialogue or presentation are converted into character strings. In the example of the automatic information providing system shown in FIG. 2, this is performed using the microphone 21 and the voice recognition device 22. Next, in step 2, the character string is divided into words to form a word string, and the part of speech of each word, the conjugation form of the conjugated word, etc. are identified. This is performed by the morphological analysis device 23 in FIG.

Next, in steps 3, 5, and 8, fields are identified, explicit topic changes are detected, and related topics are identified. Field identification in step 3 is performed as shown in Figure 4.
More on that later. This is performed by the field identification/field change detection device 24 in FIG.

If a field change is detected, proceed to steps 4 and 6. In step 4, a database is preliminarily searched for the topic, and in step 9, the database to be searched is limited to the search results. Further, in step 6, the details will be explained later, but in steps 7 and 8, variables used to hold the explicit topic and related topics are cleared.

[0016] In step 5, the linguistic expressions registered in the dictionary as shown in Fig. 3 (example of crew words used to detect an explicit topic change) are explicitly indicated by detecting them from the word string. Detect topic change points. Additionally, the start of a conversation or presentation is also detected as a topic change point. This is done by the explicit topic change detection device 25 in FIG.

[0017] If an explicit topic change is detected,
Step 6 clears the variables used to hold the explicit topic and related topics in steps 7 and 8, respectively, before proceeding to step 7. If no explicit topic change is detected, proceed directly to step 7, identifying explicit topics.

[0018] In step 7, a topic to be explicitly introduced in the process shown in FIG. 7 is identified, the details of which will be explained later. This is done by the explicit topic identification device 26 in FIG.

In step 8, topics developed in relation to the topics identified in step 7 are identified, the details of which will be explained later. This corresponds to the related topic identification device 27 in FIG.
will do.

Next, in step 9, the database limited in step 4 is searched using the current explicit topic and related topics. This is done by the database search device 28 in FIG. Finally, the information obtained by the search is provided in step 10. This is done by the information display device 29 using the display 30 and speaker 31 in FIG.

Second, the topic field identification method shown in FIG. 4 (example of topic field identification processing) will be explained. Field terms are detected from the morphological analysis results (401) using a field term dictionary such as the one shown in Figure 5 (an example of a field term dictionary used to identify topical fields) that describes the dialogue between field terms and the field to which they belong. (402), the field with the largest number of fields corresponding to the n most recently detected field terms is set as the current field. The value of n is set appropriately.

First, a FIFO (First
Prepare one In First Out) and a counter for storing the number of terms in each field for the set number of fields. In the initial state, the FIFO is empty and the value of each counter is 0. If there is a field term, the counter value of the corresponding field is incremented by 1 (403), and the set of field names is stored as one list in the FIFO (404). If, F
If the IFO is not completely filled, that is, the number of lists in the FIFO is smaller than the length n of the FIFO, the field name cannot be identified (405). If the FIFO is completely filled, the new list will be added to the FIFO.
One oldest list is taken out of the FIFO (406)
Therefore, the value of the counter corresponding to the field name included in the displayed list is decremented by 1 (407). Then, the field with the maximum counter value is set as the field at that time (408
).

Here, since field identification processing is performed using the past n field terms, there is a delay between when a field change actually occurs and when it is detected. The delay varies depending on the degree of similarity between terms in each field before and after the field change, so it is difficult to estimate the time when the topic change actually occurs. It can be done. If the degree of similarity between the fields before and after the conversion is low, the time t is valid.

[0024] When a field change occurs, explicit topics and related topics, which are more detailed topics than the field, have also changed. Therefore, as shown in Figure 6 (processing when a field change is detected), if an explicit topic change is detected after time t (601), that change will be the same as the field change detected this time. It can be considered as something that happened. Therefore, in that case, clearing of the list of explicit topics and the list of related topics in step 6 of FIG. 1 is not performed. If no explicit topic change has been detected after time t, the process of step 6 is performed (602).

Third, the explicit topic identification method shown in FIG. 7 (example of explicit topic identification processing) will be explained. First, we match a dictionary that describes linguistic expressions that explicitly introduce a topic with the results of morphological analysis of the utterance (70
1) Detect topic candidates (702). It is detected whether this topic candidate is an inappropriate word or not (703).

FIG. 8 (Example of Explicit Topic Identification Dictionary: Specific Sentence Type) is an example of a sentence pattern that explicitly introduces a topic. For example, in the sentence "X is Y.", X is detected as a topic candidate.

FIG. 9 (Example of Explicit Topic Identification Dictionary: Words Indicating Topic) is an example of a linguistic expression that explicitly introduces a topic. The dictionary in FIG. 8 is used with priority over the dictionary in FIG. 9.

[0028] If an explicit topic candidate is detected in step 703 and it is not an inappropriate word that does not represent a specific concept such as a pronoun, then the topic candidate is added to the set of explicit topics. (704). If the word is inappropriate or there are no topic candidates, nothing is done.

Related topics are identified in the same way. However, as a linguistic expression for searching for topic candidates, for example,
Use a dictionary of combinations of verbs and their surface cases, as shown in (Example of dictionary for identifying related topics). In addition, the identified related topics are managed as a set of related topics.

The processing method of the present invention will be explained in detail below along with FIG. 1, taking the dialog shown in FIG. 11 (dialogue example) as an example. The length of the FIFO for managing fields corresponding to recently detected field terms is assumed to be five.

[0031] In the initial state before the dialogue starts, the FIFO
is empty, and the counter values for each field of "postal", "aviation", and "business" are zero. Further, the sets for managing each of the explicit topics and related topics are empty.

The field identification process in step 3 of FIG. 1 will be explained. In utterance A-1, "mail" is detected as a field term. According to Figure 5, the field corresponding to this is "postal", so the value of the postal field counter is 1, and F
The state of IFO is as follows.

Since "mail" is detected as a field term in "(mail)" utterance B-1, the value of the mail field counter is 2, and the FIFO status is as follows.

[0034] “(Postal), (Postal)” In utterance A-2, “(Postal), (Postal)”
"Domestic flight" and "post" are detected as field terms in order. According to Figure 5, the field corresponding to “domestic mail” is “postal”.
and "airline", so when "domestic flight" is detected, the value of the postal sector counter becomes 3, the value of the aviation sector counter becomes 1, and the FIFO status becomes as follows.

0035 “(postal, aviation), (postal), (postal)”
Since the fields corresponding to "post" are "mail" and "business," the value of the postal field counter is 4, the value of the business field counter is 1, and the FIFO status is as follows.

[(postal, business), (postal, aviation),
(Postal), (Postal)] No field term is detected in utterance B-2.

In utterance A-3, "airmail" and "post" are detected in order. Since the field corresponding to "airmail" is "mail", the value of the mail field counter is 5, and the FIFO status is as follows.

[(Postal), (Postal, Business), (Postal, Air), (Postal), (Postal)] At this point, the FIFO
Now that the number of lists inside is equal to the FIFO length 5,
It becomes possible to identify the current field. Since the value of each field counter is 5 for postal, 1 for aviation, and 1 for business, "postal" having the maximum value is identified as the field. "
The fields corresponding to "Post" are "Postal" and "Business", so the value of the postal field counter will temporarily be 6 and the value of the business field counter will be 2, but the oldest list in FIFO (postal) is taken out from the FIFO, the value of the postal field counter is decreased by 1 to 5. Also, FIFO
The state is as below.

[(Postal, Business), (Postal), (Postal, Business), (Postal, Aviation), (Postal)] At this point, the value of the postal sector counter is the maximum, so "Postal"
is identified as a field.

In step 4 of FIG. 1, it is decided to limit the search object to only the parts related to postal services.

Regarding the detection of an explicit topic change in step 5 of FIG. 1, since the dialogue of FIG. 11 does not include a crew word as shown in FIG. 3, no detection is performed.

In the explicit topic identification in step 7 of FIG. 1, utterance A-1 in FIG. 11 is ``about (topic)'' in FIG.
Since it matches the linguistic expression "How to send mail"
is identified.

In identifying related topics in step 8 in FIG. 1, utterance A-2 in FIG. 11 is the linguistic expression "(topic) wa" in FIG. 10, and utterance A-3 is "(topic) mo" in FIG. ``Domestic flight'' and ``airmail'' match the linguistic expressions ``domestic flight'' and ``airmail.''
is identified.

In the database search in step 9 of FIG. 1, the explicit topic "How to send mail" and the related topic "Domestic mail"
Search for information related to “Airmail” and step 1
Provide it with 0.

As can be seen from the above explanation, according to this embodiment, there are three types of topics: a topic field, an explicit topic that is newly introduced, and a related topic that is developed in relation to the topic.
For each layer, the topic change point is detected using superficial linguistic expressions such as phrasing and information exchange between layers, and by identifying topics in segments where the same topic continues in each layer, it is possible to respond according to the content of the conversation. Since the appropriate information can be automatically selected and provided, it is possible to effectively promote the progress of dialogue and presentations. In other words, information related to the topic of the person conducting the dialogue or presentation is made searchable in a short period of time, so information can be provided in a short period of time during the dialogue.
The topic will develop further. Of course, it is also possible to provide information about the speaker's topic in real time.

[0046] Also, as an application field, it can also be used to support services such as information guidance services that receive inquiries from users, search large-scale databases, and promptly respond to users with the results. It is valid.

Furthermore, it can also be applied to information guidance and advertising targeted at individuals or small groups.

Although the present invention has been specifically explained above based on examples, it goes without saying that the present invention is not limited to the above-mentioned examples and can be modified in various ways without departing from the gist thereof. .

[0049]

[Effects of the Invention] As explained above, according to the present invention, it is possible to automatically select and provide appropriate information according to the content of the talk, thereby effectively promoting the progress of dialogue and presentations. can be done.

[0050] Also, as an application field, it can also be used to support services such as information guidance services that receive inquiries from users, search large-scale databases, and promptly respond to users with the results. It is valid.

Furthermore, it can also be applied to information guidance and advertisements targeted at individuals or small groups.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing the overall flow of processing for providing information in an embodiment of the automatic information providing method of the present invention.

FIG. 2 is a block diagram showing a schematic configuration of an automatic information providing system for implementing the automatic information providing method of this embodiment.

FIG. 3 is an example of a dictionary that describes linguistic expressions for detecting explicit topic changes.

FIG. 4 is a flowchart of a process for identifying topic areas.

FIG. 5 is an example of a field term dictionary used for field identification.

[Figure 6] Example of processing when a field change is detected,

[Figure 7
] An example of a flowchart of a process for identifying explicit topics.

FIG. 8 is an example of a dictionary that describes sentence patterns used to identify explicit topics.

FIG. 9 is an example of a dictionary that describes linguistic expressions used to identify explicit topics.

FIG. 10 is an example of a dictionary that describes pairs of verbs and surface cases used to identify related topics.

FIG. 11: Dialogue example.

[Explanation of symbols]

21...Microphone, 22...Speech recognition device, 23...Morphological analysis device, 24...Field identification/field change detection device, 2
5... Explicit topic change detection device, 26... Explicit topic identification device, 27... Related topic identification device, 28... Database search device, 29... Information display device, 30... Display, 31
...Speaker.

Claims (2)

[Claims] Claim 1: In an automatic information provision method that performs morphological analysis of natural language input from an input device and searches an information database based on the analysis result, a field term dictionary prepared in advance is used to clearly identify the field. An automatic method characterized by identifying topics in each of three layers of topics and topics related to the explicit topic, determining the severity of the identified topics, and searching the database for topics with high likelihood. Information provision method. 2. The automatic information providing method according to claim 1, further comprising transmitting information regarding topic changes detected in the three layers to each other.