JP4568838B2 - Visual and auditory similar product name presentation device - Google Patents

Description

  The present invention is a visual similarity (characteristic similarity that is likely to cause reading errors when reading characters) or auditory similarity (similarity that is likely to cause mistakes when listening as speech). The present invention relates to an apparatus for presenting similar product names that are easy to read or hear mistakes with respect to the product names when individual product names or the like are input as character strings by the user. This is a type of user support system that has a function to alert the user of input errors. For example, when the pharmacist is a user, it can be used as a warning system for dispensing errors due to similar drug names.

  As a means for preventing a character string input by a user from being erroneous, a technique for reducing input mistakes that occur when coding a source program is considered. In this method, the storage device stores the grammar of the language used for creating the source program, the words prepared in the above language, the input character string for creating the source program, and the like. A word satisfying a predetermined grammatical condition (for example, a function name) is extracted as a keyword from the input character string, and the search means performs a fuzzy search on the storage device, and a word similar to the keyword and When a matching word is searched and only a word similar to the keyword is found in the search means, the determination means assumes that one of the searched words is a correct word corresponding to the keyword, and the change means The keyword of the source program being created is rewritten with a word determined by the determining means to be correct (see Patent Document 1 below).

  Similarly, a character string input from the input unit is similar to this character string as a character string input method that allows the user to surely confirm an input mistake and easily correct the input character string. When a similar word in a string has a co-occurrence relationship with a word already used in the document being created, the similar word is displayed as a candidate word so that the user can confirm an input error. The thing is known (refer the following patent document 2).

  As can be seen in the above conventional example, a method for allowing a user to confirm an error in an input character string has been well known. However, in consideration of visual and auditory similarities using the distance between character shapes and the distance between phonetic symbols, what is presented to the user is not found. Alternatively, there may exist inventions having the same purpose as the present invention, but many are simple systems based on empirical knowledge, and visually and auditorily based on computer processing as in the present invention. There is no visual / auditory similar product name presentation device with an explicit configuration that evaluates and calculates similarity and incorporates a similar product name word set into the product name word encyclopedia.

  In relation to character recognition and speech recognition, calculation of the distance between words given by characters has been proposed in the past, but these are mainly as pattern recognition techniques by computer processing. In the present invention, it is also possible to make a perceptual distance such as a misreading of a character string or an erroneous listening using these techniques. However, in the present invention, as will be described later, visual and auditory In order to introduce similarity, a new distance calculation method between product name words is introduced. Further, the distance between the partial strings of the product name words can be calculated, and it is possible to deal with abbreviations such as only a part of the product name words being input.

Note that the present invention does not perform character recognition or voice recognition on a computer. Character reading or voice listening is performed by the user, and when the result is input to the computer as a character string, the product name input by the user is displayed. This is a user support system that warns that there are similar product names.
JP-A-7-160494 Japanese Patent Laid-Open No. 11-175518

  A warning about medication errors due to similar drug names is published on the website of the Japan Pharmacists Association, and a similar accident can be seen in newspaper reports. It is speculated that part of this cause is due to misreading or listening to the drug name. Similar mistakes may occur when ordering product names. In cases such as prescribing, it is usual to input the product name into the computer. In this situation, if a warning message is presented from the computer, mistakes can be reduced. An object of the present invention is to construct a high-performance system having these functions by computer processing using a large amount of data.

  The purpose is to provide a visual distance based on the character shape of the written character string of the product name word, or an auditory distance for the phonetic pronunciation of the written character string (ie, visual distance, auditory distance or both visual distance and auditory distance). ) Item name word encyclopedia in which a similar item name word set obtained in advance is calculated, a block for determining whether or not a character string input by the user exists in the dictionary item of the encyclopedia, input from the user When the entered character string does not match the item name word in the dictionary item of the encyclopedia, the block that searches for the similar item name word for the character string of the inputted item name word and the similar item name for the inputted item name word This is achieved by a similar product name presentation device having a block for presenting words.

  As preparations for constructing this device, based on a product name list prepared in advance (given by text), all product names included in this list (hereinafter referred to as “product name words”) are respectively displayed. The “perceptual distance” between the product name word pairs is calculated, and based on this, a product name word set in which the distance from the product name word is less than a certain value is obtained. As the “perceptual distance” used here, a distance corresponding to a reading error or a listening error is adopted (see embodiments <3> and <4> below). A similar product name word set for a product name word A is represented as Sn (A), n = 1, 2,. When a new product name word is added, the similar product name word set is updated by the same process.

  A measure of visual similarity (characteristic similarity that tends to cause reading errors when reading characters) and auditory similarity (similarity that can easily cause mistakes when listening as speech) of character sequences of product names, etc. When a product name or the like is input as a character string by the user, a function of issuing a warning such as an input error to the user by presenting a similar product name that is easy to read or mistaken for the product name or the like is provided. For example, when the pharmacist is a user, it can be used as a warning system for dispensing errors due to similar drug names.

  Hereinafter, embodiments of the present invention will be described.

<1> System Configuration and Operation The present invention is configured as shown in FIG. 1 and includes the following blocks.
(A) A product name word dictionary (or an electronic dictionary of that kind) incorporating a pre-calculated similar product name word set.
(A) A block for determining whether or not a character string input from a user exists in the dictionary item (corresponding to a product name word) of the encyclopedia.
(C) A block for searching for a similar product name word with respect to a character string of the input product name word (this block is activated when there is no match with the product name word input in the dictionary item of the encyclopedia).
(D) A block that presents similar product name words to the input product name words.

The operation of this device is as follows.
(1) First, when the user inputs a product name word, the character string and the dictionary item of the product name encyclopedia are collated to determine whether or not the input product name word is included in the encyclopedia. If it is included, go to (2). If not, go to (3).
(2) For the input product name word, a similar product name set is displayed as a part of the contents of the encyclopedia on the screen of a personal computer or the like used by the user. In this case, a function of notifying by voice can be added.
(3) The distance between the input product name word and the dictionary item of the product name encyclopedia is calculated, and similar product name word candidates are presented to the user. When the user selects an appropriate item name from the presented item name words, a similar item name word set for the selected item name word is presented on the screen of the personal computer together with the encyclopedia information, as in (2).

<2> Incorporation into Product Name Word Encyclopedia As preparation for constructing this device, all product names included in this list (hereinafter referred to as “Product Name Words”) based on a product name list (given as text) prepared in advance. ”), The“ perceptual distance ”(inversely proportional to the similarity) between each pair of product name words is calculated, and based on this, the distance between each product name word and itself is below a certain value. Find the product name word set. As the “perceptual distance” used here, a distance corresponding to a reading error or a listening error is adopted (see the embodiments <3> and <4> below). A similar product name word set for a product name word A is represented as Sn (A), n = 1, 2,. When a new product name word is added, the similar product name word set is updated by the same process.

  This similar product name word set is incorporated in the product name word dictionary. That is, for each item name word (dictionary item), a “list of similar item name words” is placed together with a description of the contents. When a certain product name word A is selected, “similar product name word list Sn (A), n = 1, 2,...

<3> Calculation Method of Visual Distance (Similarity) Between Product Name Words In this case, the visual distance means that an erroneous reading is likely to occur. Therefore, first, for all characters included in the article name word of the encyclopedia, the distance between the characters in the character shape (how much each character pair is similar) is calculated. This calculation itself is performed by using a feature data used in character recognition technology or the like using an actual character data sample.

  In this case, the sample data may be a typeface or a cursive form. In the case of a typeface, the distance value may be calculated from data of several standard fonts (for example, Mincho, Gothic, etc.) and the same character size. Originally, the same font as that used by the user is desirable, but this is avoided from the viewpoint of versatility. In the case of a cursive body, a large number of written character samples are required, and it is desirable to use a statistical value (for example, Mahalanobis distance, etc.) in consideration of dispersion as a distance value. When realization is difficult, a similar statistical distance may be calculated and approximated by using a large number of fonts close to cursive in a typeface font.

The distance in the shape of the character a and the character b obtained in this way is defined as df (a, b). At this time, the character strings representing the two product name words A and B are respectively
A: a1 a2 a3 ……. An
B: b1 b2 b3 ....... bm
It expresses. For example, df (a1, b1) is a distance representing the difference between the character shapes of the characters a1 and b1.

  For the perceived distance of the product name words A and B, a distance scale is introduced in consideration of the fact that the replacement of adjacent characters before and after is likely to be visually erroneous. The calculation method is as follows.

[How to calculate the visual distance between product words A and B]
(1) For each of the above character sequences, create a sequence of three groups as shown below. First about A
(a1, a2, a3), (a2, a3, a4), (a3, a4, a5),…, (an-2, an-1, an)
Similarly for B
(b1, b2, b3), (b2, b3, b4), (b3, b4, b5),…, (bm-2, bm-1, bm)
When selecting two (ai-1, ai, ai + 1) and (bj-1, bj, bj + 1) as general terms from the above series, define these two distances dff as follows: .
Distance dff {(ai-1, ai, ai + 1), (bj-1, bj, bj + 1)} = min {dff1, dff2, dff3}
In this formula, min is the minimum value of the three variables in {}, and
dff1 = {df (ai-1, bj-1) + df (ai, bj) + df (ai + 1, bj + 1)} / 3
dff2 = {df (ai-1, bj) + df (ai, bj-1) + df (ai + 1, bj + 1)} / 3
dff3 = {df (ai-1, bj-1) + df (ai, bj + 1) + df (ai + 1, bj)} / 3
It is. This definition formula means that this distance is small even if one set of three character sequences is changed. In the following, this distance is simply expressed as dff (i, j).

Given character strings A and B of two product name words, dff (i, j), i = 1,2, ... n, j = 1,2, ..., m can be calculated for two sequences . Using this dff (i, j) as a local distance, the distance between A and B is calculated using dynamic programming (hereinafter referred to as DP). The DP method used for this purpose is a method for optimizing the two sequences A and B so that they correspond best (hereinafter referred to as the “nonlinear stretch optimal matching method”). well known. Since the same method is used for the next <4>, the method will be described in <5>. Let DF (A, B) be the visual distance between A and B thus obtained. Visually similar product name words are selected based on the size of DF (A, B). At this time, it is possible to select a case where the partial character sequences are similar to each other. However, there is room for selection as to whether or not to present this.

<4> Calculation method of auditory distance (similarity) between product name words The auditory distance in this case means that it is easy to cause a mistake in hearing. In this case, it is necessary to calculate the auditory distance between the product name words. For this purpose, when the product name word is input as a character string, first, a symbol (hereinafter referred to as “speech symbol”) is used to indicate the reading. (For example, use ASCII code). As a method of giving this reading, a method used in a speech synthesis technique by a computer can be used.

  What kind of symbol system is adopted as a phonetic symbol for describing pronunciation is a problem. Here, two kinds of symbols are used. One is a symbolic system that expresses symbols known as international phonetic symbols in ASCII code (hereinafter referred to as "XSAMPA"), and the other is a sub-speech segment (Sub-Phonetic-Segment, which can be converted by rules from XSAMPA). "SPS".) From the experiments so far, it is expected that SPS is higher in terms of performance, but it takes much time to create the acoustic standard pattern described below.

The auditory distance between the phonetic symbols x and y obtained in this way is defined as dh (x, y). At this time, the phonetic symbol strings representing the pronunciation of the two product name words X and Y respectively
X: x1 x2 x3 ....... xn
Y: y1 y2 y3 ....... ym
It expresses. For example, dh (x1, y1) is the auditory distance between phonetic symbols x1 and y1, and is calculated using the acoustic standard pattern of x1 and y1.

  The auditory distance between the product name words X and Y is particularly affected by the vowel sequence in Japanese pronunciation. A distance scale that takes this into account is introduced. The calculation method is as follows.

[Method of calculating auditory distance of product name words X and Y]
As in the case of the visual distance of <3> above, the calculation method here uses the nonlinear expansion / contraction optimum matching method using DP. At this time, dh (xi, yj) is the local distance.
The auditory distance between X and Y is represented by the linear sum of the following two distances.
(1) Distance based on nonlinear expansion optimal matching between X and Y phonetic symbol sequences,
DH1 (X, Y)
(2) From the X and Y phonetic symbol sequences, only the symbol sequences related to the vowels are taken out from each of them, and the distance obtained by performing nonlinear expansion matching on the vowel symbol sequences,
DH2 (X, Y)
Using the results of (1) and (2), the auditory distance between X and Y
DH = {DH1 + αDH2} / (1 + α)
It is defined as Here, α is a positive real coefficient, and is appropriately determined by experiment. Finally, auditory similar product name words are extracted using this DH value as a scale.

<5> Nonlinear Stretching Optimal Matching Method Using Dynamic Programming (DP) This method is a method of matching two feature vector sequences by nonlinear stretching and matching so that the distance between them is minimized. It is well known in the field of speech recognition technology. Specifically, each phonetic symbol string representing the pronunciation of two product name words X and Y
X: x1 x2 x3 ....... xn
Y: y1 y2 y3 ....... ym
Is to optimize the correspondence between xi and yj, which requires local distances d (xi, xj), i = 1,2,… n, j = 1,2,…, m It becomes. Based on the correspondence optimized by DP, the distance D (X, Y) between X and Y can be calculated [Reference: Hideaki Saito et al., “Visual and auditory cognition” Ohmsha 1999, pp.198- 199].

As an extension of this technique, there is a technique called Shift Continuous DP ( Shift Continuous DP ), which can be used to calculate the optimal matching distance between any subsequences of X and Y [Reference: Ito Yoshiaki, “Performance Evaluation of Similar Space Detection Method Shift CDP and Application to Speech”, IEICE Technical Report SP 2001-36, pp. 27-34, 2001]. Basically, this method includes the above-mentioned DP by setting boundary conditions, so that more general conditions can be set.

  In this device, this shift continuous DP is used to calculate the matching distance, so that the entire character string of the product name word is not directly optimally matched, but is extracted as a similar product name word when the distance between the partial series is small. Adopting a simple algorithm. At this time, the minimum length of the subsequence is determined experimentally.

<Example of similar drug name display system based on auditory distance>
Here, the effectiveness of the present invention is shown by disclosing an example of a system that presents similar drug names based on auditory distance.

  Since there may be cases where voice is partially omitted and uttered, for the DP that is the optimum matching technique, a shift continuous DP technique that can calculate the optimum matching between the partial sequences is used.

  The name of the drug used in the experiment is approximately 4000 words of the drug name written in katakana selected from the “Drug Identification Code Encyclopedia” (Matsumitsu Takasugi supervision H15 revised edition, Pharmaceutical Journal). The drug name was converted from katakana notation to XSAMPA, SPS symbol notation which is phonetic symbol notation by the program and used.

  For the local distance dh (x, y) for the SPS symbols x and y, data already created for developing a speech recognition system was used. As the Japanese vowels, we selected 5 vowels (a, i, u, e, o, n) + repellent sounds, and selected 6 types of SPS labels corresponding to the stationary part. In the calculation of the distance DH, the coefficient α = 1.

  As a result, the distances of all combinations between the above drug names are calculated, and sample examples of drug names with a small distance are shown in the following table, for example.

  In the example of extracting a similar word in an arbitrary partial section of a long drug name, a similar drug name such as calphenyl or lorfenamine can be extracted in response to an input of “chlorpheniramine varnish cap”.

  Note that it is quite difficult to experimentally measure how well these results fit human auditory perception (especially actual users). The reason is that an actual user (in this case, a pharmacist, etc.) has preconceptions based on expert knowledge in advance and inputs a drug name etc. with a certain degree of prediction under given conditions. This is because mistakes have different conditions.

Block diagram of a visual and auditory analog name display device

Claims (2)

If the auditory distance based on the phonetic pronunciation of the character string is less than or equal to a certain value between the item name word consisting of the specific character string and the item name word having a character string different from the specific character string, the character string has the different character string A product name word dictionary in which product name words are similar product name words of product name words composed of the specific character string, and similar product name words obtained in advance for each of all product name words are incorporated,
A block for determining whether or not a character string input by a user exists in the dictionary item of the encyclopedia;
This is activated when the character string input by the user does not match the item name word in the dictionary item of the encyclopedia, and the voice between the input character string and the item name word in the dictionary item of the encyclopedia A similar product name presentation device having a block for searching for a new similar product name word whose auditory distance based on pronunciation is a predetermined value or less and a block for presenting the new similar product name word for the input product name word ,
In the auditory distance calculation, a similar item name presenting apparatus in which a distance measure for relatively evaluating the distance of a series of vowels only is introduced in a phonetic symbol string representing the reading of the item name word . The visual distance based on the difference in the arrangement of the character strings and the auditory distance based on the phonetic pronunciation of the character strings between the product name words composed of the specific character strings and the product name words having character strings different from the specific character strings. Block to calculate,
When the visual distance or the auditory distance is equal to or less than a predetermined value, the product name word having the different character string is set as a product name word similar to the product name word composed of the specific character string, and each product name word is obtained in advance. A block that creates a product name word encyclopedia that incorporates similar product name words
A block for determining whether or not a character string input by a user exists in the dictionary item of the encyclopedia;
When the character string input from the user does not match the item name word in the dictionary item of the encyclopedia, the above-described operation is performed between the input character string and the item name word in the dictionary item of the encyclopedia. A block for calculating a visual distance and the auditory distance and searching for a new similar product name word having the visual distance or the auditory distance equal to or less than a predetermined value, and the new similar product name word for the input product name word a similar product name presentation device having a block of presenting,
In the above auditory distance calculation, in the phonetic symbol string representing the reading of the product name word, a distance scale that introduces a relatively large distance of the series of vowels only is introduced ,
In the calculation of the visual distance, the visual distance between all two characters is defined based on the character shape of each character for the character describing the product name word, and between the adjacent characters in the three consecutive character strings. A similar product name presenting apparatus in which a distance scale is introduced so that alternation is evaluated to be small .

Images