JP7426302B2 - Synonym generation device and synonym generation program - Google Patents

Description

The present invention relates to a synonym generation device and a synonym generation program, and more particularly to a synonym generation device and a synonym generation program for detecting synonyms from a plurality of queries.

Computer systems that detect synonyms in databases in order to provide appropriate answers to user inquiries are known, and further research and development is underway. For example, at manufacturing or construction sites, parts may need to be replaced due to damage or wear, and in such cases, ordering replacement parts appropriately is essential for safe work at manufacturing and construction sites. is required. A similar situation may occur when ordering parts and consumables for home appliances at home.

In order to replace a part, the correct part must be selected and ordered from a huge parts database. For example, at the parts purchasing site shown in Non-Patent Document 1, it is possible to search for parts using keywords such as part names and place orders. However, when inputting keywords, search terms that are not registered in the database may be used. In this case, even if words similar to the search term are included in the database, the search will not be performed properly. This problem can be solved by incorporating a synonym dictionary into the database, but manually creating a synonym dictionary is costly and time consuming.

For this reason, a system for automatically creating a synonym dictionary is provided by, for example, Non-Patent Document 2. In this system, in two sentences to be compared, if the target word pair is different from each other, but all surrounding words are the same, the target word pair is determined to be synonymous. and register it in the synonym dictionary.

for example,
・"My eyes hurt, so I'm going to the eye doctor tomorrow."
・"My eye hurts so I'm going to the eye doctor tomorrow."
Suppose there are two sentences.
These two sentences have different pairs of words: "ophthalmologist" and "eye doctor." On the other hand, the words that appear around these word pairs are all the same: "me,""eyes,""ouch,""tomorrow," and "go." Therefore, "ophthalmology" and "eye doctor" can be determined to be synonymous and can be registered in the synonym dictionary.

In contrast,
・"My eyes hurt, so I'm going to the eye doctor tomorrow."
・Suppose there are two sentences: ``I have a toothache, so I'm going to the dentist tomorrow.''
In these two sentences, the word pair "ophthalmologist" and "dentist" is found, but there are also different words surrounding this word pair, "eye" and "teeth." Therefore, it can be determined that "ophthalmology" and "dentist" are not synonyms, and are excluded from registration in the synonym dictionary.

However, when synonyms are automatically registered using the method described in Non-Patent Document 2, there is a problem that words that are not synonyms may be erroneously registered as synonyms. Furthermore, when inputting a query sentence, the user may make an incorrect input, or in the case of voice input, a mistake may occur. In this case as well, a word that is misspoken may be mistakenly input as a synonym.

Hitachi Global Life Solutions, Hitachi home appliance consumable parts direct online sales “Parts Shop”, https://parts.hitachi-cm.com/pshop/, viewed 2020/03/23 Zellig S. Harris, “Distributional Structure,” WORD, vol.10, no.2-3, pp.146-162, 1954.

Japanese Patent Application Publication No. 2013-016011

An object of the present invention is to provide a synonym generation device and a synonym generation program that can efficiently generate a synonym dictionary while avoiding erroneous registration of synonyms.

In order to solve this problem, the synonym generation device of the present invention includes an inquiry record table that records inquiries from user input/output devices, a synonym candidate table that registers synonym candidates that are synonym candidates, It has a synonym dictionary storage unit that stores a synonym dictionary for recording words, and a control unit that determines whether registration in the synonym dictionary is possible and updates the synonym dictionary. The control unit registers the synonym candidate in the synonym candidate table based on the inquiry recorded in the inquiry record table, and performs synonym determination regarding the synonym candidate registered in the synonym candidate table. is executed, and synonym candidates that meet predetermined criteria are registered as synonyms in the synonym dictionary.

Further, the synonym generation program according to the present invention includes a step of storing an inquiry from a user input/output device in an inquiry record table, and a synonym that becomes a synonym candidate based on the inquiry recorded in the inquiry record table. The computer is configured to perform the steps of detecting candidates and registering them in a synonym candidate table, and storing synonym candidates that meet predetermined criteria as synonyms in a synonym dictionary storage unit.

According to the present invention, it is possible to provide a synonym generation device and a synonym generation program that can efficiently generate a synonym dictionary while avoiding erroneous registration of synonyms.

FIG. 1 is a block diagram illustrating a synonym generation device according to an embodiment. 3 is a flowchart showing the overall flow of database inquiry processing executed in the embodiment. 12 is a flowchart illustrating another example of the overall flow of database inquiry processing executed in the embodiment. 3 is a flowchart illustrating details of the execution procedure of the database inquiry process (step S20) in FIG. 2. FIG. 4 is a schematic diagram showing an example of a data structure of an inquiry record table 42. FIG. It is a flowchart further explaining the procedure of input processing 20B. It is a flowchart explaining a specific example of SQL generation execution processing (step S20C). 5 is a schematic diagram showing an example of a parts database 52. FIG. It is a flowchart explaining the details of a synonym detection process (step S30). 4 is a schematic diagram showing an example of the data structure of a synonym candidate table 44. FIG. 10 is a flowchart illustrating detailed steps of the synonym registration process (step S30E) in FIG. 9. FIG. 5 is a schematic diagram showing an example of a data structure of a synonym dictionary database 50. FIG. It is a conceptual diagram explaining the principle of determining a lower limit value. 14 is a flowchart illustrating an example of a procedure for determining a lower limit value of the number of synonym candidate detections according to the principle illustrated in FIG. 13. This is an example of a display screen of the user input/output device 8. This is an example of a display screen of the user input/output device 8.

This embodiment will be described below with reference to the accompanying drawings. In the accompanying drawings, functionally similar elements may be designated by the same number. Although the attached drawings show embodiments and implementation examples in accordance with the principles of the present disclosure, they are for the purpose of understanding the present disclosure, and should not be used to limit the present disclosure in any way. isn't it. The descriptions herein are merely typical examples and do not limit the scope of claims or applications of the present disclosure in any way.

Although the embodiments are described in sufficient detail for those skilled in the art to implement the present disclosure, other implementations and forms are possible without departing from the scope and spirit of the technical idea of the present disclosure. It is necessary to understand that it is possible to change the composition and structure and replace various elements. Therefore, the following description should not be interpreted as being limited to this.

A computer system (synonym generating device) according to an embodiment will be described with reference to FIG. FIG. 1 is a diagram showing the configuration of a computer system 1 according to an embodiment. This computer system 1 includes a CPU (Central Processing Unit) 2, a main storage device 3, an external storage device 4, and an input/output device 5, and is configured by interconnecting these devices.

The computer system 1 can be connected to a user input/output device 8 via a computer network 7 . The user input/output device 8 is configured to be able to inquire about parts to the computer system 1 and obtain the search results. The user input/output device 8 is, for example, a mobile phone such as a smartphone, a tablet PC (Personal Computer), a notebook PC, a desktop PC, etc., but is not limited to these. The user input/output device 8 includes, for example, an input device such as a keyboard and a microphone, and an output device such as a display, and is not limited to a specific format.

The CPU 2 is a central control device that controls the entire computer system 1 according to a computer program, and constitutes a control section of the computer system 1 together with the main storage device 3 and the external storage device 4. The main storage device 3 is a storage device that stores modules and tables that constitute the main part of the operation of the computer system 1. Further, the external storage device 4 is a storage device for executing operations of the computer system together with the main storage device 3, and mainly functions as a database.

The input/output device 5 is configured to receive various data input by an operator and to display various data (computation results, input data, received data) in order to appropriately execute calculations in the computer system 1. The input/output device 5 may include, for example, a keyboard, a mouse, a display, a microphone, a camera, a speaker, etc., but is not limited to a particular type.

Specifically, the main storage device 3 stores, for example, a database inquiry processing module 20, a synonym detection processing module 30, an inquiry record table 42, a synonym candidate table 44, etc., and also stores a synonym candidate detection frequency lower limit value. I remember 46. The database inquiry processing module 20, synonym detection processing module 30, inquiry record table 42, and synonym candidate table 44 collectively constitute a synonym generation program.

The database inquiry processing module 20 receives an inquiry from the user input/output device 8 (for example, an inquiry including a model name, part name, and part name), processes the inquiry, and sends the search results to the user input/output device 8. It has the function of returning. The synonym detection processing module 30 is configured to perform a process of detecting synonyms from input inquiry processing according to a sequence described below.

The inquiry record table 42 is a table that records inquiries input from the database inquiry processing module 20. As will be described later, the inquiry record table 42 includes data indicating the content of each of a plurality of inquiries and whether the search was successful or unsuccessful (whether a search result was obtained or not). There is. Further, the synonym candidate table 44 is a table that stores synonym candidates extracted from various inquiries as words (candidates) that can be synonyms.

In this embodiment, words determined to be potentially synonymous according to predetermined rules are not immediately registered in the synonym dictionary database 50, but are registered in the synonym candidate table 44 once. . Thereafter, in accordance with the process described later, a determination process is performed on these synonym candidates, and synonym candidates that meet a predetermined determination criterion are officially registered and updated in the synonym dictionary database 50 as synonyms. The predetermined criteria will be described in detail later, but for example, one of the criteria may be whether or not the same synonym candidate has been detected a number of times exceeding the lower limit 46 of synonym candidate detection times.

The external storage device 4 stores a synonym dictionary database 50 (synonym dictionary storage section) and a parts database 52. The synonym dictionary database 50 is a database in which synonyms detected by the synonym detection processing module 30 are registered. The synonym dictionary database 50 may hold a certain amount of synonym data from the time of shipment of the computer system 1, for example, or may not have synonym data at the time of shipment, but synonyms that are accumulated as a result of use. It may also be something that records word data. The parts database 52 is a database that stores names and numbers of parts in association with corresponding model names and part names.

When there is an inquiry from the user input/output device 8, the database inquiry processing module 20 searches the parts database 52, and if a part matching the inquiry is found, returns the result to the user input/output device 8. During the search, the database inquiry module 20 also searches the synonym dictionary database 50, and also searches whether the name of the part or part related to the inquiry exists in the synonym dictionary database 50. If a synonym is found as a result of the search, the search is executed by replacing the word that does not exist in the parts database 52 with the synonym.

FIG. 2 is a flowchart showing the overall flow of database inquiry processing executed in this embodiment. When an inquiry regarding parts is input from the user input/output device 8 to the computer system 1 via the computer network 7, the database inquiry processing module 20 operates and database inquiry processing is executed (step S20).

Thereafter, a synonym detection process is executed to analyze the input inquiry and detect synonyms from among the words included in the plurality of inquiries (step S30). Details of the synonym detection process (step S30) will be described later. In step S40, it is determined whether the overall process is to be ended. If not (No), the process returns to the database inquiry process (step S20); otherwise (Yes), the entire process is ended. The determination in step S40 can be made by displaying on the user input/output device 8 or the like whether or not the process is to be terminated, and determining whether or not the process is to be terminated based on a specific input signal from the user input/output device 8 or the like. This determination is not limited to any particular method or machine.

Another example of overall processing will be described with reference to FIG. In this example, after step S20, step S25 includes a process of determining whether or not to perform synonym detection processing. If Yes, the process moves to synonym detection processing (step S30); otherwise (No), the process returns to database inquiry processing (step S20). The determination in step S25 may be made by displaying on the user input/output device 8, etc. whether or not to execute the synonym detection process, and making a Yes/No determination according to a specific input signal from the user input/output device 8, etc. can.

Further, although not shown, the following operations may be added to the flow of FIG. 2 or 3. The following operations can be combined as appropriate. It goes without saying that the operations are not limited to these operations.
・When the database inquiry process (step S20) is executed a predetermined number of times, the process proceeds to the synonym detection process (step S30). ・When it is detected that there is no input from the user input/output device 8 for a certain period of time, the synonym is detected. Execute detection processing - Execute synonym detection processing when it is detected that CPU 2 is in an idle state (if CPU 2 is busy, synonym detection processing is not executed and database inquiry processing is repeated) )
- Detects the time from computer system 1, etc., and if the detected time is night (for example, in the range of 8:00 PM to 5:00 AM), executes synonym detection processing. - If a successful query was detected in the previous database query process. Perform synonym detection processing when

Next, details of the database inquiry process (step S20) will be explained with reference to the flowchart of FIG. In the database inquiry process, when an inquiry is input, one entry is created in the inquiry record table 42 (step S20A).

FIG. 5 is a diagram showing an example of the data structure of the inquiry record table 42. FIG. 5(a) shows the data structure of the inquiry record table 42 immediately after creating the first entry. In FIG. 5A, a one-entry table with three fields has been created, including, from the left, a serial number field for an inquiry, a field for a list of words included in the inquiry, and a field for success/failure of the inquiry. A "successful inquiry" means that one or more search results are returned by an inquiry to the parts database 52, and a failure means that there are no search results. In the following, a successful inquiry will be referred to as a "successful inquiry", and an unsuccessful inquiry will be referred to as a "failure inquiry".

Returning to FIG. 4, the explanation will be continued. After step S20A, input processing (step S20B) is executed. In the input process (step S20B), sentences input from the keyboard, microphone, etc. are processed so that they can be easily handled in the next SQL generation execution process (step S20C). Details of the input processing (step S20B) will be described later.

Next, SQL generation execution processing (step S20C) is executed. In the SQL generation/execution process (step S20C), an SQL query is generated from the input sentence, the SQL query is executed against the parts database 52, and the success/failure of the query is recorded. Details of the SQL generation execution process (step S20C) will be described later.

In the following step S20D, it is determined whether the result of the SQL generation execution process (step S20C) was successful or unsuccessful. If the inquiry is unsuccessful (No), the inquiry is recorded as a "failed inquiry" (step S20E). For example, for a "failure query", set a serial number starting from 1 and incrementing by 1 in the serial number field of an empty entry in the query record table 42, the word used for the SQL query in the word list field, and False in the query success record field. can be recorded by recording.

FIG. 5(b) shows the data structure when 1 is recorded as the serial number, "ZX200, hoist, screw" is recorded as the word list, and False is recorded as the inquiry success record in the first entry of the inquiry record table 42. There is. Once such recording is done, the process returns to step S20A.

Further, FIG. 5C shows an example of the entry creation state of the inquiry record table 42 when the process returns to step S20A. A new empty entry is created above the entry with serial number 1.

If the determination in step S20D is successful (Yes), the process moves to step S20F, the inquiry is recorded as a "successful inquiry" in an empty entry in the inquiry record table 42, and the database inquiry processing 20 ends. Recording as a "successful query" means that the serial number starting from 1 and increasing by 1 is entered in the serial number field of the empty entry in the query record table 42, the word used for the SQL query is entered in the word list field, and "True" is recorded in the query success record. ” is to be recorded. FIG. 5(d) shows a case where n is recorded as the serial number, “ZX200, crane, bolt” is recorded as the word list, and True is recorded as the inquiry success record in the nth entry (n is a positive integer) of the inquiry record table 42. It shows.

Although the inquiry record table 42 in FIG. 5 is in the form of a table, it is not limited to this form, and may also be implemented in a form such as a linked list or a hash table.

The procedure of the input process (step S20B) will be further explained with reference to the flowchart of FIG. 6. FIG. 6(a) is a flowchart when the input is a keyboard input. In step S20B1, a sentence input from the keyboard is subjected to morphological analysis to extract nouns. Morphological analysis is an analysis method that extracts words from natural language sentences and classifies them into nouns, particles, etc. For example, in morphological analysis, the nouns "ZX200," "crane," and "bolt" are extracted from the natural language sentence "ZX200, crane, bolt." Note that the punctuation marks are inserted to make the explanation easier to understand, and are not necessary for morphological analysis. FIG. 6(b) is a flowchart when the input is a voice input. The audio may be the result of microphone input, or an audio file in an appropriate format may be used as input. In step S20B2, speech text conversion is performed on the input speech. Speech-to-text conversion is called Speech to Text technology, and can be performed on smartphones and smart speakers such as Amazon Echo (registered trademark) and Google Home (registered trademark). In the following step S20B3, morphological analysis is performed on the text-converted sentence to extract nouns.

In the above example of input processing (step S20B), keyboard input and voice input are taken as examples, but in addition to these, there are other methods such as recognition of character strings included in images from still images and videos, sign language, etc. A variety of input methods are possible, such as photographing the gesture with a camera as input, displaying predetermined words and sentences as options, and letting the user select one.

A specific example of the SQL generation execution process (step S20C) will be described with reference to the flowchart in FIG. In step S20C1, the nouns extracted in the input process 20B are set as noun 1, noun 2, etc., and an SQL query as shown below is generated, for example.
"SELECT part number FROM parts DB WHERE model name=noun 1 AND part name=noun 2 AND part name=noun 3"
For example, if the extracted nouns are "ZX200", "crane", and "volt",
The SQL query will be "SELECT part number FROM parts DB WHERE model name=ZX200 AND part name=Crane AND part name=Bolt".

FIG. 8 is a diagram showing an example of the parts database 52. From the left, the first column stores the model name, the second column stores the part name, the third column stores the part name, and the fourth column stores the part number. The SQL query in step S20C1 in FIG. 7 has a form that reflects the configuration of this parts database 52. If there is a change in the configuration of the parts database 52, or if there is a change in the information (column) to be retrieved, the corresponding SQL query will also be changed. It is also conceivable that the order of the input nouns is not necessarily in the order of the conditions of the WHERE clause in the SQL query. In such a case, an SQL query can be created and executed using all combinations of conditions such as model name, part name, etc. and input nouns. It is also possible to create a model name dictionary, part name dictionary, etc. in advance, and if the noun is a word registered in the model name dictionary, it can be applied to the model name condition (noun 1). The query generation method is not limited to these methods. Additionally, if synonym data already exists in the synonym dictionary database 50, check whether there are any synonyms for Noun 1, Noun 2, etc., and if there are synonyms, replace them with the synonyms before creating the SQL query. You can also do it.

Returning to FIG. 7, the explanation will be continued. In step S20C2, the generated SQL query is executed against the parts database 52. Next, in step S20C3, if one or more search results are obtained by the SQL query executed on the parts database 52, the query is determined to be a successful query (True). On the other hand, if there are no search results, the inquiry is determined to be a failed inquiry (False).

Details of the synonym detection process (step S30) will be described with reference to the flowchart in FIG. When the synonym detection process starts, the value of the inquiry success record field of the top (nth) entry in the inquiry record table 42 is True, and the inquiry of the entry one below (n-1st) It is determined whether the value of the success record field is False (step S30A). If so (Yes), the process moves to step S30B, and if not (No), the synonym detection process ends. In step S30B, n is assigned to variable k. Additionally, a synonym candidate table 44 is created.

FIG. 10 is a diagram showing an example of the data structure of the synonym candidate table 44. As an example, the synonym candidate table 44 has first to third columns in order from the left side.

The first column on the left records parts database terms, which are terms registered in the parts database 52. The second column in the middle records synonym candidates forming a synonym pair with the parts database term. Further, the third column records the number of synonym candidate detections, which is the number of times this synonym pair was detected.

In the first row of the synonym candidate table 44 in FIG. 10, the term "screw" is registered as a synonym candidate for the term "bolt" registered in the parts database 52. This means that the number of times the term "screw" has been detected as a synonym candidate for "bolt" (synonym candidate detection) is 135 times. In addition, in the synonym candidate table in FIG.
The term ``bottle'' is registered as a candidate synonym for the term ``bottle''. As will be described later, this "bottle" indicates a case where the user should have inputted "volt" by voice or keyboard, but instead inputted "bottle" by mistake. Therefore, for (volt, bottle), the number of synonym candidate detections is set to one (few users make such a mistake).

Note that in the synonym candidate table 44, both the first column and the second column are empty when the system is shipped (before the start of use), and "0" may be recorded in the third column. As synonym candidates are registered and detected multiple times, the number in the third column also increases.

Returning to FIG. 9, the explanation will be continued. In step S30C, the word lists of the k-th entry and the k-1-th entry in the inquiry record table 42 are compared to detect synonyms. If the words registered in the word list are the same in the k-th entry and the k-1-th entry except for one word, those one words are treated as a synonym pair. Among the synonym pairs, the word included in the k-th entry is a component database term, which is a word already registered in the component database 52, and the word included in the k-1th entry is defined as a synonym (synonym candidate). For example, if the word list of the k-th entry is "ZX200, crane, bolt" and the k-1st word list is "ZX200, crane, screw" (corresponding to the case where k=n in Figure 5(d) ), "bolt" and "screw" are a synonym pair, "bolt" is a term in the parts database 52, and "screw" is a synonym (synonym candidate). If two or more words are different in both entries, it is assumed that no synonyms were detected.

Next, in step S30D, it is checked whether synonyms have been detected in both entries. If a synonym has been detected (Yes), the synonym registration process (step S30E) is executed, and if it has not been detected (No), the synonym detection process (step S30) is ended. Details of the synonym registration process (step S30E) will be described later.

Next, in step S30F, k-1 is substituted for k. That is, a query that was originally a failed query is regarded as a successful query, and is compared with another failed query to perform synonym detection processing. This point will be discussed later. Next, in step S30G, it is checked whether k>0, and if k>0 (Yes), the process returns to step S30C, otherwise (No), the process proceeds to step S30H, where the inquiry record table 42 is deleted and synonyms are detected. The process (step S30) ends.

As shown in steps S30B to S30G, if synonyms can be detected from the set of success queries and failure queries, the serial number of the entry is decremented (reduced) by one, and the failure queries that differ by one word from the success queries are Search for. If such a failed inquiry is found, the failed inquiry can be treated as a successful inquiry from now on, and the operations in the flowchart of FIG. 9 can be executed. In other words, a query that was originally recognized as a failed query can be compared with other failed queries and used for synonym detection. Explain why this is possible.

The word list for the success inquiry (corresponding to the n-th entry in FIG. 5(d)) is composed only of words registered in the parts database 52. On the other hand, as a result of comparison with the successful query, the word list of the failed query for which synonyms were detected (corresponding to the n-1 entry in FIG. It consists only of the synonyms given. Therefore, by replacing the detected synonyms with the corresponding words registered in the parts database 52, this failed inquiry can be rewritten to the same word list as the successful inquiry. Therefore, this failed inquiry can be regarded as a successful inquiry, and an attempt can be made to see if a synonym can be detected between it and the previous failed inquiry, for example. By repeating this process recursively until no synonyms can be detected or until the first entry of the inquiry record table 42 is reached, synonyms can be detected efficiently.

It is also possible to detect synonyms by matching successful queries with all failed queries and repeat the process recursively, but it is clear that such an algorithm will cause combinatorial explosion, resulting in a large amount of computation time and It is not practical in terms of the computer resources required for calculation. This concludes the explanation of the synonym detection procedure.

The detailed procedure of the synonym registration process (step S30E) in FIG. 9 will be described with reference to the flowchart in FIG. 11. First, in step S30E1, if the synonym candidate table 44 has not yet been created, it is created in the main storage device 3. Even if there is a synonym candidate table 44, if there is no free entry in the synonym candidate table 44, a free entry is added. The fields for synonyms of parts database terms in empty entries are initialized to blanks, and the synonym detection count field is initialized to 0.

Next, in step S30E2, words that may be synonymous with the term in the parts database 52 detected in step S30C of FIG. Increment the count by 1 (if it was originally 0, set it to 1).

Next, in step S30E3, a lower limit value of the number of synonym candidate detections is determined. The number of times a word is detected as a possible synonym is counted, and if the count value exceeds this lower limit, the word that is considered as a synonym candidate is finally recognized as a synonym. , is registered in the synonym dictionary database 50. Details of step S30E3 for determining the lower limit value of the number of synonym candidate detections will be explained using FIGS. 12 and 13.

Next, in step S30E4, it is checked whether the number of synonym candidate detections after incrementing in step S30E2 is larger than the lower limit value 46 of the number of synonym candidate detections determined in step S30E3, and if Yes, step S30E5 is executed. , No, the synonym registration process ends. In step S30E5, words that may be synonymous with the detected parts database term are finally registered in the synonym dictionary database 50 as synonyms. Then, entries in the synonym candidate table 44 in which synonyms of the parts database terms are registered are deleted. At this time, if there is no free entry in the synonym dictionary database 50, a free entry is created and registered. However, if the parts database term and the synonym have already been registered as a pair in the synonym dictionary database 50, no redundant registration will be performed.

FIG. 12 shows an example of the data structure of the synonym dictionary database 50. For example, in the synonym dictionary database 50, terms included in the parts database 52 are registered in a first column, and synonyms corresponding to terms included in the parts database 52 are registered in a second column.

Details of step S30E3 (determining the lower limit value of the number of synonym candidate detections) in FIG. 11 will be described with reference to FIGS. 13 and 14. FIG. 13 is a conceptual diagram illustrating the principle of determining the lower limit value. FIG. 14 is a flowchart illustrating details of the operation of step S30E3.

The horizontal axis of the graph 90 in FIG. 13 indicates the number of synonym candidate detections, and the vertical axis indicates the number of pairs of synonyms and synonym candidates (number of synonyms) having the same number of synonym candidate detections. The graph (distribution diagram) of the number of synonym candidate detections and the number of synonyms is expected to be a bimodal distribution with two peaks, as shown in graph 90. This is because synonyms that are likely to be entered by many people will have a higher number of synonym candidate detections, but since each person makes mistakes when saying or inputting them, it is expected that the number of synonym candidate detections will be lower. This is because it will be done. Therefore, the number of synonym candidate detections corresponding to the bottom of the bimodal distribution of the graph 90 can be determined to be the lower limit value 46 of the number of synonym candidate detections. The flowchart in FIG. 14 shows an example of a procedure for determining the lower limit of the number of synonym candidate detections based on such a principle.

First, in step S30E3A, the lower limit value of the number of synonym candidate detections is tentatively determined to be the maximum value that the variable can take. For example, if the lower limit of the number of times of synonym detection is a 32-bit signed integer, 2147483647 can be provisionally determined as the maximum value.

Next, in step S30E3B, it is checked whether the number of entries in the synonym candidate table 44 is sufficiently large. For example, it is possible to check whether the number of entries is sufficiently large by checking whether it is larger than a predetermined threshold (for example, 100). The threshold value can be provided externally in the form of a configuration file or the like, or can be directly coded into the program, but is not limited to a specific method. If it is determined that it is equal to or greater than the threshold value (Yes), the process proceeds to step S30E3C, and if not, the synonym candidate detection frequency lower limit determination step S30E3 is ended.

In step S30E3C, a graph (see FIG. 13) is created in which the two axes are the number of synonym candidate detections (horizontal axis) in the synonym candidate table 44 and the corresponding number of synonyms (vertical axis). Then, in step S30E3D, the minimum value of the graph is found, and the number of times of synonym candidate detection corresponding to the minimum value with the smallest number of times of synonym candidate detection is set as the lower limit value 46 of the number of times of synonym candidate detection. Thereby, the synonym candidate detection frequency lower limit determination step S30E3 is completed.

By performing the process shown in FIG. 14, as long as the number of entries in the synonym candidate table 44 is small (less than or equal to the integer given in advance), the lower limit value of the number of synonym candidate detections can be indicated by the variable. Since this is the maximum value, the condition of step S30E4 in FIG. 11 is not always satisfied (determined as No), and no synonym is registered from the synonym candidate table 44 to the synonym dictionary database 50.

On the other hand, after the number of entries becomes sufficiently large and the bimodal distribution of the graph in FIG. 13 becomes clear, the value of the lower limit value 46 of the number of synonym candidate detections is set based on the graph. The condition in step S30E4 is now satisfied (determined as Yes), and synonyms are registered from the synonym candidate table 44 to the synonym dictionary database 50. This method allows synonym dictionary registration according to the characteristics of the input. Note that an algorithm other than the one shown in FIG. 13 can also be used for finding the bottom of the bimodal distribution. Further, the lower limit value 46 of the number of synonym candidate detections can be given in advance from outside in the form of a configuration file or the like, without performing the processing shown in FIG.

FIG. 15 is an example of a display screen of the user input/output device 8. As shown in FIG. This display screen includes, for example, a Q&A screen 8A, a selection result display screen 8B, and an option display screen 8C.

The Q&A screen 8A displays the results of using the computer system 1 interactively. The option display screen 8C displays conversion candidates for the voice-to-text conversion performed in step S20B2 (FIG. 6) described above. In addition, if the parts database 52 has columns other than model name, part name, part name, and part number (for example, machine number), and the part number cannot be uniquely identified only by the model name, part name, and part name, information Columns whose information is unknown (for example, a list of machine numbers) can also be displayed on the option display screen 8C. It is also possible to display other options. These options are selected by tapping the screen or clicking on the relevant location with a mouse, and the selection results are displayed on the selection result display screen 8B. Note that, instead of the user input/output device 8, the input/output device 5 can also be used to perform similar displays and operations.

FIG. 16 shows display examples of the Q&A screen 8A, the selection result display screen 8B, and the option display screen 8C. Note that for the purpose of explaining the operation, the lower limit value 46 of the number of synonym candidate detections will be explained as being set to 0.

On the Q&A screen 8A, as shown by reference numeral 8A in FIG. 16A, questions necessary for specifying a part number, such as model name, part name, and part name, are displayed. In response, assume that the user inputs an inquiry such as "ZX〇〇" and "screw for hoist" by text input or voice input. By inputting such a query, an SQL query is generated and an inquiry to the parts database 52 is executed.

In this case, since the part name "lifter" and the part name "screw" are not registered in the parts database 52, the search result for the part number is 0 (failure inquiry 1).

Next, change the wording a little and enter the query ``crane screw.'' However, since the component name "screw" is not registered in the component database 52, the search results are still 0 (failed inquiry 2).

Next, by changing the wording further and inputting "crane bolt", the search in the parts database 52 hits, and one search result "Part number B1234" is returned (Success Inquiry 1). Further, it is displayed that the parts database term "bolt" and the synonym "screw" and the parts database term "crane" and the synonym "lifter" have been detected. This is because by comparing success query 1 and failure query 2, it is detected that screw and bolt are synonyms (if the procedure in Figure 11 is also executed), and furthermore, failure query 2 is considered to be a success query. A comparison with failure inquiry 1 shows that crane and hoist are considered synonymous.

Note that if a plurality of candidates are found as a search result and the part number is not specified, another question may be displayed. For example, a question regarding the purchase history of the part desired to be identified, or other phrase suggestions based on the search history may be displayed. This increases the number of synonym registrations and improves the accuracy of search results.

A selection result display screen 8B shown in FIG. 16 shows an example of a final confirmation screen displayed when one of a plurality of options is specified. In this figure, part "XX" is selected, and if you want to finalize it, select "Yes", and if you want to make a new selection, select "Back". Note that, when a plurality of part numbers are hit, a list of candidates is displayed on the option display screen 8C, and the user is prompted to select from the plurality of candidates.

Although the above embodiment has been described using the search for the parts database 52 as an example, synonyms can be detected using the same method when searching for blueprints, maintenance documents, and the like. Furthermore, in the above embodiment, it is assumed that the parts database 52 is a relational database, and an SQL query is generated to perform a search. words can be detected.
Furthermore, the software and the like that realize each of the above-mentioned functional units can be recorded on a portable magnetic or optical recording medium, or can be installed on a computer using them. Furthermore, it is also possible to install it on a computer by downloading it via a network such as the Internet.

The present invention is not limited to the embodiments described above, and includes various modifications. For example, the above-described embodiments have been described in detail to explain the present invention in an easy-to-understand manner, and the present invention is not necessarily limited to having all the configurations described. Furthermore, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. Furthermore, it is possible to add, delete, or replace some of the configurations of each embodiment with other configurations.

1...Computer system (synonym generator)
2...CPU
3...Main storage device 4...External storage device 5...Input/output device 7...Computer network 8...User input/output device 20...Database inquiry processing module 30...Synonym detection processing module 42...Inquiry record table 44...Synonym candidate table 46 ...Synonym candidate detection count lower limit 50...Synonym dictionary database 52...Parts database

Claims (4)

an inquiry record table for recording inquiries from user input/output devices;
a synonym candidate table that registers synonym candidates that are synonym candidates;
a synonym dictionary storage unit that stores a synonym dictionary that records synonyms;
a control unit that determines whether registration in the synonym dictionary is possible and updates the synonym dictionary;
The control unit registers the synonym candidate in the synonym candidate table based on the inquiry recorded in the inquiry record table,
performing synonym determination regarding the synonym candidates registered in the synonym candidate table, and registering synonym candidates that meet a predetermined determination criterion as synonyms in the synonym dictionary ;
The control unit registers the synonym candidate as a synonym in the synonym dictionary based on a comparison result between the number of detections detected as the synonym candidate and a lower limit value,
The control unit creates a graph of the number of detections and the number of synonyms, which is the number of pairs of synonyms detected that number of times, detects a minimum value of the graph, and sets the lower limit based on the minimum value. set value
A synonym generating device characterized by: an inquiry record table for recording inquiries from user input/output devices;
a synonym candidate table that registers synonym candidates that are synonym candidates;
a synonym dictionary storage unit that stores a synonym dictionary that records synonyms;
a control unit that determines whether registration in the synonym dictionary is possible and updates the synonym dictionary;
The control unit registers the synonym candidate in the synonym candidate table based on the inquiry recorded in the inquiry record table,
performing synonym determination regarding the synonym candidates registered in the synonym candidate table, and registering synonym candidates that meet a predetermined determination criterion as synonyms in the synonym dictionary ;
The inquiry record table is capable of recording the contents of the inquiry and success/failure of search results based on the inquiry,
The control unit detects the synonym candidate by comparing a word included in a successful query in which the search result is success and a word included in an unsuccessful query in which the search result is failure;
The synonym candidate is detected by comparing a failed query in which a synonym was detected with other failed queries.
A synonym generating device characterized by : storing an inquiry from a user input/output device in an inquiry record table;
Detecting synonym candidates that are synonym candidates based on the inquiry recorded in the inquiry record table, and registering them in the synonym candidate table;
and storing synonym candidates that meet predetermined criteria as synonyms in a synonym dictionary storage unit,
The step of storing the synonym candidate as a synonym in the synonym dictionary storage section includes storing the synonym candidate in the synonym dictionary storage section based on a comparison result between the number of detections detected as the synonym candidate and a lower limit value. registered as a synonym in the department,
The step of storing the synonym candidate as a synonym in the synonym dictionary storage unit involves creating a graph of the number of detections and the number of synonyms, which is the number of pairs of synonyms detected that number of times, and detecting a minimum value of and setting the lower limit value based on the minimum value;
Synonym generator. storing an inquiry from a user input/output device in an inquiry record table;
Detecting synonym candidates that are synonym candidates based on the inquiry recorded in the inquiry record table, and registering them in the synonym candidate table;
and storing synonym candidates that meet predetermined criteria as synonyms in a synonym dictionary storage unit,
The inquiry record table is capable of recording the contents of the inquiry and success/failure of search results based on the inquiry,
detecting the synonym candidate by comparing words included in a successful query in which the search result is success and words included in a failed query in which the search result is failure;
Detecting the synonym candidate by comparing a failed query in which a synonym has been detected with other failed queries;
Synonym generator.

Images