JP6608119B2 - Drug name output device, drug name output method, and drug name output program - Google Patents

Description

  The present invention relates to a drug name output device, a drug name output method, and a drug name output program for receiving a drug name search target character string and outputting a drug name by referring to a data storage unit using the character string. Is.

  Conventionally, character string search has been widely used. In many cases, the user manually inputs a character string as a search query and obtains an answer from the search server. At that time, it is assumed that the character string of the search query includes some ambiguity such as a spelling error but does not include a large amount or a large error.

  The slight ambiguity as described above can be eliminated by a neighborhood search using, for example, normalization processing or edit distance, and an appropriate search can be performed. The normalization process is a process for converting the search query into a predetermined unified format. For example, the conversion of endings such as conversion from “go” to “go” and conversion from “apples” to “apple” Processing, processing to correct the shaking of the notation that converts “diamond” to “diamond”, unified processing to uppercase or lowercase that converts “Apple” to “apple”, and “apple” to “apple” There is a unified process to convert to half-width or full-width.

  In the neighborhood search using the edit distance or the like, the word similarity is measured based on the edit distance or the like, and a search result including a word close to the search query is output. For example, if the edit distance is allowed up to “1”, a set of words with one character difference can be obtained from the search query.

JP 2012-133539 A Japanese Patent Laid-Open No. 10-240870 Japanese Patent Laid-Open No. 7-152871

  Here, in the future, it is assumed that input methods (images or sounds) other than the manual input as described above will be widened. In addition, there is a possibility that meaningful search results cannot be obtained including errors and noises that cannot be caused by humans. In particular, advertisements and packaging include special typefaces, designs or logos, and it is difficult to perform OCR appropriately. As a result, for example, if an AND search is performed on a recognition result including noise, there is no corresponding result, and a large amount of meaningless results are obtained if an OR search is performed on the recognition result including noise.

  The normalization process described above is based on the premise that there are no errors in the search query that are large and cannot be made by humans, and therefore an appropriate search cannot be performed even if the OCR result or the like is used for the search.

  Further, in the neighborhood search using the edit distance described above, the size of the set (number of words) increases exponentially with respect to the edit distance, and the time required for the search also increases. Simple application is difficult if the error location cannot be assumed to be limited, such as OCR results.

  Patent Document 1 proposes a method for realizing a fuzzy search by using the OCR result including noise as a search target, calculating a feature vector by clustering search character strings, and collating with the feature vector of the search target. Yes.

  However, in the method of Patent Document 1, since information other than the input character string and the search character string is required to suppress the influence of noise on the feature vector generation, it is difficult to suppress the influence of noise.

  In Patent Document 2, the character string of the OCR result is divided for each word, the number of appearances of the word is totaled, and the validity of the recognition result is checked based on the number of appearances, thereby improving the word recognition rate. However, it is a method for improving the accuracy of OCR in a document including a large number of words, and it is difficult to perform an appropriate search when the search character string based on the OCR result includes a lot of noise. Further, Patent Document 3 proposes to improve the character recognition rate of a character string as a result of OCR by combining two different character recognition results and determining with a basic tree dictionary. It is difficult to perform an appropriate search when a search character string based on the result includes a lot of noise.

  In view of the above problems, the present invention provides a drug name output device, a drug name output method, and a drug name output program capable of outputting an appropriate drug name in a search based on a character string including noise in the drug name. The purpose is to provide.

  The drug name output device of the present invention is a search unit that searches for a drug name by referring to a data storage unit in which a drug name is stored using a search target character string, and a search result searched by the search unit. The number of parts in which any of the partial character strings generated by dividing the character string representing one of the plurality of drug names matches the search target character string, and the partial characters similar to each of the partial character strings The appearance frequency determined according to the number of parts that match the column is calculated, and the reliability based on the appearance frequency is calculated for each of the plurality of drug names in the search results, and each drug name is trusted. And an evaluation unit for calculating the degree.

  The search unit performs a search using a combination of partial character strings obtained by dividing the search target character string as a search query.

  In addition, the search target character string is a character string acquired by OCR, and similar partial character strings are fitted with other characters estimated for characters that are likely to be erroneously recognized by OCR among the partial character strings. New substring.

  The drug name output device of the present invention further includes an additional search necessity determination unit that determines whether or not an additional search is necessary based on the reliability calculated by the evaluation unit, and the search unit includes the additional search. A search result integration unit that performs an additional search based on the search target character string and merges and integrates the search result of the first search and the result of the additional search when the necessity determination unit determines that an additional search is necessary. Can be further provided.

  Further, the evaluation unit arranges in order from the drug name of the search result in which the appearance frequency of the partial character string is high, and if the number of drug names of the search result is larger than a preset threshold, It is possible to exclude the drug names in the search results of the ranks from the search results of the first search.

  In addition, the additional search necessity determination unit acquires the drug name of the search result with the highest occurrence frequency of the partial character string, and the ratio that the partial character string of the acquired drug name is included in the search target character string is determined in advance. If it is larger than the set threshold value, it is determined that an additional search is not necessary, and if the ratio is equal to or less than the threshold value, it can be determined that an additional search is necessary.

  Further, the search query based on the search target character string used for the initial search may be different from the search query based on the search target character string used for the additional search.

  In the medicine name output method of the present invention, the medicine name output device searches the medicine name by referring to the data storage unit storing the medicine name using the search target character string, and a plurality of search results are obtained by the search. The number of parts where any one of the partial character strings generated by dividing the character string representing the drug name of any one of the drug names matches the search target character string, and the partial character string similar to each of the partial character strings The frequency of appearance determined in accordance with the number of portions that match with each other, and processing for calculating the reliability based on the frequency of appearance is performed on each of the plurality of drug names in the search result, and the reliability of each drug name And drug names are output in descending order of reliability.

  The drug name output program of the present invention includes a search unit that searches a computer for a drug name with reference to a data storage unit that stores drug names, and a plurality of drug names that are search results searched by the search unit. The number of parts where one of the partial character strings generated by dividing the character string representing any one drug name matches the search target character string, and the part that matches the partial character string similar to each of the partial character strings An evaluation for calculating the reliability of each drug name by obtaining the frequency of appearance determined according to the number of drugs and calculating the reliability based on the frequency of appearance for each of a plurality of drug names in the search result And an output unit that outputs drug names in descending order of reliability.

  According to the medicine name output device, medicine name output method, and medicine name output program of the present invention, an appropriate medicine name can be output from a short character string including noise in the medicine name.

The figure which shows schematic structure of the search system using one Embodiment of the character string search apparatus of this invention. The flowchart for demonstrating the effect | action of the search system using one Embodiment of the character string search apparatus of this invention. Flow chart for explaining the first search result evaluation method Diagram for explaining matching of search target character string and search result character string The figure which shows an example of PTP (Press Through Package) read photoelectrically The figure which shows the example of a display of a search result at the time of applying the search system shown in FIG. 1 to a chemical | medical agent test | inspection system.

  Hereinafter, a search system using an embodiment of a character string search apparatus and method and a program of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of a search system according to the present embodiment.

  As shown in FIG. 1, the search system of this embodiment includes a character string search device 1, an OCR (Optical Character Recognition) device 2, a database device 3 (corresponding to a data storage unit of the present invention), and a display 4. And.

  The character string search device 1 is obtained by installing an embodiment of the character string search program of the present invention on a computer.

  The character string search device 1 includes a CPU (Central Processing Unit), a semiconductor memory, a hard disk, and the like, and an embodiment of the character string search program of the present invention is installed in the hard disk. Then, the character string search program is executed by the control unit 17 having a CPU, whereby a character string receiving unit 10, a search unit 11, an evaluation unit 12, an additional search necessity determination unit 13, and a search as shown in FIG. The result integration unit 14, the storage unit 15, and the output unit 16 operate.

  The character string accepting unit 10 accepts a character string to be searched. Specifically, the character string receiving unit 10 according to the present embodiment receives a character string that is a result of OCR output from the OCR device 2. The OCR device 2 performs character recognition processing on image data obtained by photoelectrically reading a reading target including a character string, thereby recognizing and outputting the character string included in the image data. As a character recognition process, a known process has already been performed. However, in this character recognition process, a symbol, mark or pattern other than a character is erroneously recognized, and noise is included in the result of the actual character recognition process. Yes. In the present embodiment, it is assumed that a considerably long character string including a large amount of miscellaneous noise due to such misrecognition is received by the character string receiving unit 10.

  The search unit 11 performs a search by inputting the character string received by the character string receiving unit 10 and referring to data stored in the database device 3 using the character string. The search unit 11 of the present embodiment does not directly use the character string received by the character string receiving unit 10, that is, the result of OCR, but generates a search query from this character string and uses the search query to create a database device. The data stored in 3 is searched. As a method for generating a search query, for example, a character string can be divided into partial character strings, and AND of the partial character strings can be used as a search query. As a character string dividing method, for example, N-Gram division can be used. The search query generation method is not limited to the method described above, and other known methods can be used.

  As the search algorithm in the search unit 11, a known search algorithm can be used.

  The evaluation unit 12 receives the character string of the search result first searched by the search unit 11 and the character string of the search target received by the character string reception unit 10, and the character string of the input search result and the search target The character string of the search result is evaluated by collating with the character string. A method for evaluating the search result character string will be described later.

  The additional search necessity determination unit 13 determines whether or not an additional search of the character string to be searched is necessary based on the evaluation result of the character string of the search result in the evaluation unit 12. This additional search is an additional search based on the search target character string received by the character string receiving unit 10. Therefore, a search is performed so that a search result different from the search performed first is obtained.

  Specifically, for example, the search may be performed using a search query different from the search query used for the initial search, or a search algorithm different from the search algorithm used for the initial search may be used. As a search query different from the search query used for the initial search, for example, a search query having a different ambiguity from the search query used for the initial search can be used. More specifically, for example, when the character string to be read in the OCR is “Ashita” and the search query used for the first search is “Ashi AND Shita”, the search query for the additional search is “Ashi OR”. “Shita” can be used. The search query generation method with different ambiguities is not limited to this, and other known methods can be used.

  As a search algorithm different from the search algorithm used for the first search, for example, a search algorithm that refers to a search history can be used.

  A method for determining whether or not an additional search is necessary will be described later.

  The determination result in the additional search necessity determination unit 13 is output to the search unit 11, and the search unit 11 performs the additional search described above when the determination result indicates that the additional search needs to be performed. If the determination result is not necessary, the search result of the first search is output to the storage unit 15 and stored without performing the additional search described above.

  The search result integration unit 14 merges and integrates the search result of the first search in the search unit 11 and the search result of the additional search described above. The additional search described above is performed when the evaluation unit 12 evaluates that the search result of the first search is not sufficient, but the search result of the additional search is not always better than the first search result. Therefore, in this embodiment, the search result of the first search and the search result of the additional search are merged and integrated so that the first search result is also effectively used without being discarded. In this merger and integration, if the search result of the first search and the search result of the additional search overlap, the duplication is excluded.

  The storage unit 15 includes a storage device such as a semiconductor memory and a hard disk. The search result of the first search in the search unit 11, the search result merged and integrated in the search result integration unit 14, and the evaluation in the evaluation unit 12 The result is stored.

  The output unit 16 reads out the search results of the first search stored in the storage unit 15, the merged and integrated search results, the evaluation results, and the like, and outputs them to the display 4 for display.

  As described above, the control unit 17 has a CPU, and controls the entire character string search device 1 based on the character string search program of the present embodiment.

  The database device 3 stores data related to the character string to be searched. Specifically, for example, various drug names are stored. Then, the medicine package on which the medicine name is printed is photoelectrically read and the character string recognized by the OCR device 2 is used as a character string to be searched, and collated with various medicine names stored in the database device 3. And a search is performed. The data stored in the database device 3 is not limited to the drug name, and various data are stored depending on the application.

  Next, the operation of the search system of this embodiment will be described with reference to the flowchart shown in FIG.

  First, the character string recognized by the character recognition process is output from the OCR device 2 as a character string to be searched and accepted by the character string accepting unit 10 (S10).

  The search target character string received by the character string reception unit 10 is output to the search unit 11, and the search unit 11 generates a search query from the input search target character string and uses the search query to create a database device. The first search is performed by referring to the data stored in 3 (S12).

  Then, the first search result in the search unit 11 is output to the evaluation unit 12, and the evaluation unit 12 evaluates the input search result. The search result is evaluated in this way because the OCR result includes noise, so the search query used in the initial search may not be appropriate, that is, the initial search result may not be appropriate. is there. Hereinafter, the first search result evaluation method will be described in detail with reference to the flowchart shown in FIG. FIG. 4 is a diagram for explaining the collation between the search target character string and the search result character string in the evaluation of the first search result.

  First, the evaluation unit 12 acquires a search target character string received by the character string reception unit 10, that is, a result of OCR in a state including noise, and sets j = 0 (S30). j is 0 ≦ j <M, and M is the number of first search results.

  Next, the evaluation unit 12 takes out S [j] (0 ≦ j <M), which is one of the first search results, and performs normalization processing and N-Gram character string conversion on the S [j]. Division processing is performed to generate a partial character string (S32). The number of partial character strings generated at this time is N, each partial character string is P [i] (0 ≦ i <N), and i = 0 is set. The counter r is set to r = 0, and the counter f is set to f = 0.

  The normalization process is a process of aligning the search target character string or search query into a preset unified format, and as described above, the ending change is converted into a unified format or the notation fluctuation is corrected. Or unification processing to uppercase or lowercase, unification processing to half-width or full-width.

  Then, the evaluation unit 12 takes out the partial character string P [i] (S34), determines whether there is a portion that matches P [i] in the character string to be searched acquired in S30, If there is a matching part, the counter r is set to r = r + 1 (S36). In this determination, it may be determined whether there is not only a portion matching P [i] but also a portion matching a partial character string group similar to P [i]. Specifically, in the OCR device 2, for example, when erroneous recognition of “si” and “tsu” is likely to occur, not only the partial character string “ashi” but also “ashi” and “atsu” are partial character strings. What is necessary is just to determine whether there exists a part which corresponds to these partial character strings as a group. Hereinafter, generation of such a partial character string group is referred to as character string expansion.

  Next, the evaluation unit 12 measures the number of portions that match P [i] in the character string to be searched (S38). If the number of portions that match P [i] is c, then f = f + c. At this time, character string expansion may be performed as in S36.

  Then, i = i + 1 is set, and it is determined whether i <N. If i <N, the processes from S34 to S40 are repeated (S42, YES). That is, it is determined whether or not all partial character strings P [i] are included in the character string to be searched, r is counted, and P [i] is included in the character string to be searched. The number of matching parts is measured and f is counted.

  When it is determined that i <N is not satisfied in S42, the process proceeds to S44, and the evaluation unit 12 calculates the ratio of the partial character string included in the character string to be searched. Specifically, R [j] = r / N is calculated as an index related to S [j], and is stored in the storage unit 15.

  Furthermore, the evaluation unit 12 calculates the appearance frequency of the partial character string included in the character string to be searched (S46). Specifically, F [j] = f / N is calculated as an index related to S [j], and is stored in the storage unit 15.

  Then, the evaluation unit 12 sets j = j + 1, determines whether j <M, and if j <M, repeats the processing from S32 to S48 (S50, YES). That is, R [j] and F [j], which are evaluation indexes, are calculated for all the first search results S [j], and are stored in the storage unit 15. In the present embodiment, R [j] and F [j] correspond to the reliability in the present invention.

  On the other hand, if j <M is not satisfied in S50, the process proceeds to S52, and the evaluation unit 12 arranges S [k] (0 ≦ k <M), which is the first search result, in descending order of F [k]. That is, the evaluation unit 12 arranges S [k] in descending order of F [k], which is an appearance frequency index. Then, for example, when the number M of the first search results is larger than a preset threshold value, the evaluation unit 12 excludes lower search results having a relatively small value of F [k] from the search results. This narrows down the first search result. Note that the number of search results to be excluded is set in advance by the user. The evaluation unit 12 stores the narrowed search result in the storage unit 15.

  The above is the description of the evaluation method of the search result of the first search in the evaluation unit 12.

  Next, returning to FIG. 2, in S14, after the evaluation of the first search result in the evaluation unit 12 is completed as described above, the additional search necessity determination unit 13 determines whether the above-described additional search is necessary. (S16). Specifically, the additional search necessity determination unit 13 is the character string of the search result having the highest rank among S [k] arranged in descending order of F [k] in S52 shown in FIG. R [m] related to [m] is read from the storage unit 15, and the R [m] is compared with a preset threshold value. If R [m] is larger than the threshold value, it is determined that no additional search is necessary ( S16, NO). In this case, as described above, the search result obtained by narrowing down the first search result is read from the storage unit 15 and output to the output unit 16. It should be noted that, instead of determining whether or not an additional search is necessary based only on the appearance frequency index F [k], the additional search is performed in consideration of R [m], which is an index of the proportion of partial character strings. This is because, if it is determined only by F [k], only the appearance frequency of some partial character strings may be high. By considering R [m], it can be confirmed whether most of the partial character strings are included in the character string to be searched.

  On the other hand, the additional search necessity determination unit 13 determines that an additional search is necessary when R [m] is equal to or smaller than the threshold (S16, YES). When the additional search necessity determination unit 13 determines that an additional search is necessary, a message to that effect is output to the search unit 11, and the search unit 11 performs the additional search as described above. The result is stored in the storage unit 15 (S18).

  By performing the search in two stages in this way, the occurrence of redundant search results can be suppressed in the initial search, and a wider search result can be obtained in the additional search.

  After the additional search, the search result integration unit 14 reads out the search result of the additional search and the narrowed-down initial search result from the storage unit 15, merges and integrates them, and stores the merged and integrated search result as the storage unit. 15 (S20).

  Next, the merged and integrated search result is read by the evaluation unit 12, and the evaluation unit 12 sets the partial character string as a search target in the same manner as the evaluation of the first search result described above. Evaluate by matching against a string.

  Then, for example, F [j] is calculated for each character string of the search result merged and integrated, and each character string is scored based on this.

  Each character string and its score are stored in the storage unit 15, and then read out by the output unit 16. For example, the character strings and their scores are arranged in descending order and displayed on the display 4. At this time, all the character strings of the merged and integrated search results may be displayed side by side, or only the character string having the highest score may be displayed. Only the character string of the score may be displayed.

  Next, an example in which the above-described search system is applied to a drug inspection system will be described. The medicine inspection system is a system for inspecting whether or not a medicine prescribed to a patient is correct according to a doctor's prescription.

  Specifically, in this medicine inspection system, a character string including a medicine name, a pharmaceutical company name, a medicine amount, and the like printed on a medicine packaging delivered to a patient is photoelectrically read, and the read image Data is input to the OCR device 2 of the search system. Examples of the medicine packaging include PTP (Press Through Package). FIG. 5 shows an example of PTP read photoelectrically. In FIG. 5, only the drug name is printed at a plurality of locations. However, as described above, in some cases, the name of the pharmaceutical company, the amount of the drug, or a mark may be printed.

  Then, in the OCR device 2, character recognition processing is performed on the image data, and a character string included in the image data is acquired. If drug names are printed at a plurality of locations in the image data as shown in FIG. 5, these character strings may be connected to perform a search as one character string, or individual character strings You may search as

  The acquired character string is input to the character string search device 1, and the character string search device 1 performs the above-described search, evaluation, additional search, and the like using the input character string. At this time, data of a large number of drug names are stored in the database device 3, and the character string search device 1 collates the input character string with a large number of drug names stored in the database device 3. Perform searches and additional searches.

  And as above-mentioned, a score is calculated about the search result by the character string search device 1, for example, as shown in FIG. 6, a medicine name is displayed from the top in order with a high score. The patient name is set and inputted in advance according to the doctor's prescription for the drug testing system, and the drug name is compared with the drug name with the highest score in the search result in the drug testing system. It is determined whether or not the medicine passed to is correct. When the medicine is correct, the test OK is displayed. When the medicine is not correct or when there is no searched medicine, display such as prompting for visual confirmation is performed. Note that only the display of the drug name of the search result and the display of the drug name of the prescription may be performed, and the determination may be performed by the user.

  In addition, when applying the search system of the said embodiment to a chemical | medical agent test | inspection system as mentioned above, it is good also considering the database apparatus used for an additional search as a database apparatus different from the database apparatus used for the first search, for example. Specifically, database devices having different types of medicines stored in advance may be used.

  Moreover, in the search system of the said embodiment, although the character string reception part 10 received the character string output from the OCR apparatus 2, it is not restricted to an OCR apparatus, For example, the character string acquired by the speech recognition process is used. You may make it acquire. That is, any character string may be used as long as it includes noise such as a character string acquired by OCR or voice recognition. In such a character string, the effect of the present invention is remarkably exhibited. Of course, the character string accepting unit 10 may accept a character string that does not include noise.

  In the description of the search system of the above embodiment, the additional search is performed once. However, the additional search may be performed twice or more based on the evaluation result of the merged and integrated search result.

DESCRIPTION OF SYMBOLS 1 Character string search apparatus 2 OCR apparatus 3 Database apparatus 4 Display 10 Character string reception part 11 Search part 12 Evaluation part 13 Additional search necessity judgment part 14 Search result integration part 15 Storage part 16 Output part 17 Control part

Claims (9)

A search unit that searches for a drug name with reference to a data storage unit in which the drug name is stored using a search target character string;
Number of portions in which any one of the partial character strings generated by dividing a character string representing one drug name of a plurality of drug names as search results searched by the search unit matches the search target character string A frequency of appearance determined according to the number of parts matching a partial character string similar to each of the partial character strings, and calculating a reliability based on the appearance frequency, An evaluator that applies each of the drug names to calculate the reliability of each of the drug names;
A drug name output device comprising: an output unit that outputs drug names in descending order of reliability.   The drug name output device according to claim 1, wherein the search unit performs a search using a combination of partial character strings obtained by dividing the search target character string as a search query. The search target character string is a character string obtained by OCR;
The said similar partial character string is a new partial character string which applied another character estimated with respect to the character which is easy to produce the misrecognition by OCR among the said partial character strings. Drug name output device. An additional search necessity determination unit that determines whether an additional search is necessary based on the reliability calculated in the evaluation unit;
When the search unit determines that the additional search is necessary in the additional search necessity determination unit, performs the additional search based on the search target character string,
The medicine name output device according to any one of claims 1 to 3, further comprising a search result integration unit that merges and integrates the search result of the first search and the result of the additional search.   When the evaluation unit arranges in order from the drug name of the search result in which the appearance frequency of the partial character string is high, and the number of drug names of the search result is larger than a preset threshold, the preset number 5. The drug name output device according to claim 4, wherein drug names in the search results of a lower rank are excluded from search results of the first search.   The additional search necessity determination unit acquires the drug name of the search result with the highest appearance frequency of the partial character string, and the partial character string of the acquired drug name is included in the search target character string The drug name according to claim 5, wherein when the ratio is larger than a preset threshold, it is determined that the additional search is unnecessary, and when the ratio is equal to or less than the threshold, it is determined that the additional search is necessary. Output device.   The drug name output according to any one of claims 4 to 6, wherein a search query based on the search target character string used for the first search is different from a search query based on the search target character string used for the additional search. apparatus. The drug name output device
Using the search target character string, the drug name is searched with reference to the data storage unit in which the drug name is stored,
The number of parts in which any one of the partial character strings generated by dividing a character string representing one drug name of a plurality of drug names as a search result by the search matches the search target character string, and the part A process of obtaining an appearance frequency determined according to the number of parts that match a partial character string similar to each of the character strings, and calculating a reliability based on the appearance frequency is performed for each of the plurality of drug names in the search result. To calculate the reliability of each of the drug names,
A drug name output method, wherein drug names are output in descending order of reliability. Computer
A search unit that searches for a drug name with reference to a data storage unit in which the drug name is stored using a search target character string;
Number of portions in which any one of the partial character strings generated by dividing a character string representing one drug name of a plurality of drug names as search results searched by the search unit matches the search target character string A frequency of appearance determined according to the number of parts matching a partial character string similar to each of the partial character strings, and calculating a reliability based on the appearance frequency, An evaluator that applies each of the drug names to calculate the reliability of each of the drug names;
A drug name output program that functions as an output unit that outputs drug names in descending order of reliability.