JP2017142640A - Similar document recommendation system, similar document recommendation method, and similar document recommendation program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable selection of one with the same purpose with an API relating to a service from an API relating to another service at a short time.SOLUTION: A recommendation device 5 comprises an API information holding unit 53 which holds a plurality of pieces of API information including sentences explaining a plurality of first APIs, an input unit which accepts input of an API document having a sentence explaining a second API, a similarity calculation unit 59 which calculates the similarity between the API document and the API information, and an API recommendation unit 54 which recommends one or a plurality of first APIs similar to the second API on the basis of the similarity calculated by the similarity calculation unit 59.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a similar document recommendation system, a similar document recommendation method, and a similar document recommendation program.

  With the recent increase in network speed and the spread of web browsers, APIs (Application-Program Interface) that provide predetermined processing via the Internet have become widespread. This API is a rule that defines the procedure and data format for calling and using programs and management data on a server via the Internet from other servers and terminals. It is called by imitating an API that is a convention for calling and using.

  Developers can easily develop applications for end users by using APIs. In addition, API providers can appeal their services and content to end users, etc., using applications that utilize this API. For example, a communication carrier can promote development of an application for an end user by using a network function or an operation function as an API, and can appeal its communication service to an end user or the like.

  In recent years, devices such as smartphones have diversified, and the market has become global, making it easier for businesses to enter new niche areas. Under such circumstances, it is important to provide APIs in a form that can secure profits. In order to realize these services and secure profits, it is desirable that APIs (mashups) be easy between operators. To enable API linkage, it is desirable to define API technical standards and unify them into standard APIs that conform to them.

  For example, Non-Patent Document 1 describes this standard API model. TM (TeleManagement) FORUM, the publisher of this Non-Patent Document 1, is working on the examination of industry standards, the promotion of commercialization based on that, and the activation of the industry in order to realize network management of interoperable information communication systems. It is a non-profit industry organization established for the purpose, and is examining a standard API for providing services across multiple operators.

TeleManagement FORUM, “Frameworx Specification Product Ordering API REST Specification”, Release 14.5.0, November 2014

  For service A for which an API is provided by a service provider or the like, it may be desired to match API_X of service X with API_A of service A. In other words, there is a demand for standardizing API specifications and unifying them into standard APIs. As a result, the interoperability of the service A is improved, and the service A can be provided across more network operators and cloud operators. For this purpose, it is conceivable to select and recommend an API document to be used for the same purpose from the API group of service A for the input service X API document.

  However, in order to standardize APIs, it is necessary to select APIs to be used for the same purpose, but at present, it takes a lot of time to manually read API specification documents and match each API.

  An object of the present invention is to solve the above-described problems and to select an API having the same purpose as an API related to a service from APIs related to different services in a short time.

  In the first aspect of the present invention, a holding unit that holds a plurality of API information having a sentence that describes a plurality of first APIs, an input unit that receives an input of an API document having a sentence that describes a second API, and the API A similarity calculator that calculates the similarity between the document and the API information, a recommendation unit that recommends one or more first APIs similar to the second API based on the similarity calculated by the similarity calculator, A similar document recommendation system characterized by comprising:

  By doing in this way, the thing with the same purpose as 2nd API which concerns on a service can be selected from 1st API which concerns on a different service in a short time.

  The invention according to claim 2, wherein the similarity calculation unit corrects a sentence included in the API document using teacher data, and the similar document recommendation system according to claim 1, did.

  In this way, it is possible to normalize the terms that are notating and improve the accuracy of similarity calculation.

  In the invention according to claim 3, the API document has a request parameter example and / or a response parameter example, and the input unit deletes the request parameter example and the response parameter example included in the API document. The similar document recommendation system according to claim 1, wherein:

  In this way, it is possible to delete the request parameter example and the response parameter example that cause noise in the document vector, and improve the accuracy of similarity calculation.

  In the invention according to claim 4, the API document includes a character string or a word including a number, and the input unit deletes the character string or the word including the number included in the API document. The similar document recommendation system according to claim 1, wherein:

  In this way, it is possible to delete the character string or word including a number in the request parameter example that causes noise in the document vector, and improve the accuracy of similarity calculation.

  In the invention according to claim 5, the API document includes a camel case or snake case character string, and the input unit uses the camel case or snake case character string included in the API document as a word. The similar document recommendation system according to claim 1, wherein

  By doing so, the character string of the camel case or snake case that becomes the noise of the document vector can be decomposed into words that become the constituent elements of the document vector, and the accuracy of the similarity calculation can be improved.

  According to a sixth aspect of the present invention, in the similar document recommendation system according to the first aspect, the recommendation unit stores a user's correctness / incorrectness determination for the recommendation result of the first API.

  By doing in this way, the correctness determination result is reflected in the next similarity calculation, so that the accuracy of the similarity calculation can be gradually improved.

  In the invention according to claim 7, a step of holding a plurality of API information having a sentence explaining a plurality of first APIs, a step of receiving an input of an API document having a sentence explaining a second API, the API document, A similar document recommendation method comprising: calculating a similarity with the API information; and recommending one or more first APIs similar to the second API based on the similarity .

  By doing in this way, the thing with the same purpose as 2nd API which concerns on a service can be selected from 1st API which concerns on a different service in a short time.

  In the invention according to claim 8, holding means for holding a plurality of API information having a sentence explaining a plurality of first API, input means for receiving an input of an API document having a sentence explaining a second API, the API document, The computer functions as a similarity calculation unit that calculates a similarity with the API information, and a recommendation unit that recommends one or more first APIs similar to the second API based on the similarity calculated by the similarity calculation unit. This is a similar document recommendation program.

  By doing in this way, the thing with the same purpose as 2nd API which concerns on a service can be selected from 1st API which concerns on a different service in a short time.

  According to the present invention, an API having the same purpose as an API related to a service can be selected from APIs related to different services in a short time.

It is a block diagram of the system in this embodiment. It is a figure which shows the detail of a recommendation apparatus. It is a figure which shows an API document example. It is a flowchart which shows a document process process and a similarity calculation process. It is a figure which shows the objective of API commonization. It is a block diagram which shows common API conversion by the mediator of a comparative example.

  Next, modes for carrying out the present invention (referred to as “embodiments”) will be described in detail with reference to the drawings as appropriate.

By standardizing and unifying API specifications, service interoperability is improved, and services can be provided across more network operators and cloud operators. On the other hand, existing services may already have their own APIs that do not conform to the common API, and the service provider system or the network provider system needs to be modified to make it a common API.
5 (a) and 5 (b) are diagrams showing the purpose of API sharing, and FIG. 5 (a) is a diagram showing a state before API sharing.
The service provider system 1a provides services to the network provider system 2a through API_X, provides services to the network provider system 2b through API_Y, and provides services to the cloud provider system 2c through API_Z. That is, these API_X, API_Y, and API_Z are different from each other. That is, it is difficult to connect the service provider system 1a with the network provider systems 2a and 2b and the cloud provider system 2c.

FIG. 5B is a diagram showing a state after API sharing.
The service provider system 1a provides services to the network provider systems 2a and 2b and the cloud provider system 2c by API_A. Further, the other service provider system 1b also provides services to the network provider systems 2a and 2b and the cloud provider system 2c by API_A. In other words, this system has a common API.
By sharing the API, the service provider system 1a can be easily connected to the network provider systems 2a and 2b and the cloud provider system 2c when expanding the service. Other service provider systems 1b and the like can easily enter a service. In addition, each service provider system 1a, 1b can easily acquire a new customer.
Furthermore, network operators and loud operators can reduce costs by using commercially available products.
However, in order for the service provider system 1a and the network provider system 2a to modify the existing processing unit so that API_X is shared with API_A, it takes a large number of migration steps. Therefore, an API conversion device called a mediator (see FIG. 6 to be described later) is often used for common API.

FIG. 6 is a configuration diagram showing common API conversion by the mediator 4 of the comparative example.
The mediator 4 includes an API conversion unit 45, API input units 41 and 44, API output units 42 and 43, a scenario input unit 46, and an API conversion scenario 47. The user registers the API conversion scenario 47 in advance via the scenario input unit 46. The network operator system 2 requests an API of the API_X specification related to the existing service X. The service provider system 1 provides an API of API_A specifications related to the common service A. The mediator 4 converts API_X into API_A based on the API conversion scenario 47 and outputs it to the service provider system 1, and converts API_A into API_X and outputs it to the network provider system 2.

  The network operator system 2 transmits a command based on the API_X specification to the service operator system 1 via the mediator 4. The API input unit 44 receives the API request transmitted from the network operator system 2, and the API conversion unit 45 changes the API request to the API_A specification based on the API conversion scenario 47. The API output unit 43 outputs the converted API request to the service provider system 1.

The service provider system 1 transmits the API response to the network provider system 2 after executing processing based on the API request. The API response transmitted by the service provider system 1 is transmitted to the network provider system 2 via the mediator 4.
The API input unit 41 receives the API response of the API_A specification transmitted from the service provider system 1, and the API conversion unit 45 changes the API response to the API_X specification based on the API conversion scenario 47. The API output unit 42 outputs the converted API response.
As a result, the API_X specification network operator system 2 can operate with the API_A specification service operator system 1. That is, the network operator system 2 is converted into a common API by the mediator 4.

  In implementing the mediator 4, it is necessary to create an API conversion scenario 47 in advance and input it via the scenario input unit 46. At this time, it is necessary for the user of the mediator 4 to match the API_X specifications that are existing services and match each API included in the API_A specifications, and the API conversion scenario 47 requires a lot of time and man-hours. The recommendation device according to the present embodiment supports the generation of the API conversion scenario 47 by automatically matching APIs in order to save the creation of the API conversion scenario 47.

FIG. 1 is a configuration diagram of a system in the present embodiment.
The mediator 4 includes an API conversion unit 45, API input units 41 and 44, API output units 42 and 43, a scenario input unit 46, and an API conversion scenario 47. The user receives an API recommendation from the recommendation device 5 and creates an API conversion scenario 47, and registers the created API conversion scenario 47 using the scenario input unit 46 of the mediator 4.
The recommendation device 5 includes an API document input unit 51, an API matching unit 52, an API information holding unit 53, and an API recommendation unit 54.
The API document input unit 51 receives the API document 6 input by the user, and the API matching unit 52 matches the API. The API recommendation unit 54 recommends an API to the user. The API information holding unit 53 holds information related to the common API. The recommendation device 5 will be described in detail with reference to FIGS. 2 and 3 to be described later.

  The network operator system 2 requests an API of the API_X specification related to the existing service X. The service provider system 1 provides an API of API_A specifications related to the common service A. The mediator 4 converts the API request of the API_X specification into the API_A specification and outputs it to the service provider system 1, and converts the API response of the API_A specification into the API_X specification and outputs it to the network operator system 2.

The network operator system 2 transmits a command based on the API_X specification to the service operator system 1 via the mediator 4. The API input unit 44 receives the API request transmitted from the network operator system 2, and the API conversion unit 45 changes the API request to the API_A specification based on the API conversion scenario 47. The API output unit 43 outputs the converted API request to the service provider system 1.
The service provider system 1 transmits the API response to the network provider system 2 after executing processing based on the API request. The API response transmitted by the service provider system 1 is transmitted to the network provider system 2 via the mediator 4.

  The API input unit 41 receives the API response of the API_A specification transmitted from the service provider system 1, and the API conversion unit 45 changes the API response to the API_X specification based on the API conversion scenario 47. The API output unit 42 outputs the converted API response.

FIG. 2 is a diagram showing details of the recommendation device 5.
In addition to the API document input unit 51, API matching unit 52, API information holding unit 53, and API recommendation unit 54 shown in FIG. 1, the recommendation device 5 further includes a correct / incorrect determination result unit 55 and a teacher data holding unit 56. Composed. The API document input unit 51 includes a receiving unit 57 and a processing unit 58. The API matching unit 52 includes a similarity calculation unit 59. The API information holding unit 53 holds API information that is a specification of API_A (first API) related to the service A as a comparison destination in advance. This API information includes sentences describing a plurality of APIs.
When the user inputs the entire text of the API document to be compared with the recommendation device 5 and the parameter information, the recommendation device 5 calculates the similarity between the API documents and recommends the API to be used for the same purpose to the user. .

The accepting unit 57 is a part of the API document input unit 51, and an API document 6 (API) that describes one or more API specifications among API_X (second API) specifications related to the service X input by the user. Information). In FIG. 2, the API document 6 is described as an API document.
The receiving unit 57 allows two types of reception of the entire sentence of the API document 6 and a sentence having a request parameter and a response parameter. However, reception of the entire text of the API document 6 is essential. When the entire sentence of the API document 6 is input, the API recommendation is performed and the parameter is not recommended. The APIs that the user inputs to the reception unit 57 may be input individually one by one, or a certain API group may be input collectively. The accepting unit 57 can accept the API document 6 including a request parameter example and a response parameter example.

  The processing unit 58 is a part of the API document input unit 51, and by deleting a part in which the request parameter example and the response parameter example are described in the API document 6, the selection accuracy of the API used for the same purpose is increased. ing. This is because the request parameter example and the response parameter example are specific expressions of the API document 6 and become noise for the calculation of the existing document similarity. Specifically, the processing unit 58 determines and deletes a portion described in JSON (JavaScript (registered trademark) Object Notation) format or XML (eXtensible Markup Language) format and a character string of a word including a number. Judgment of JSON format and XML format is performed by finding the part enclosed in braces {} and angle brackets <>. By the processing of the processing unit 58, noise in the calculation of the similarity calculation unit 59 can be removed.

  The similarity calculation unit 59 calculates the similarity between the API document 6 and each API information held by the API information holding unit 53. The similarity calculation unit 59 uses the synonym set of the industry-specific terminology 561 held in the teacher data holding unit 56 to replace words, thereby absorbing the fluctuation of the service-specific expression and improving the accuracy of API selection. . Furthermore, the similarity calculation unit 59 presents all API groups applicable to the category as candidates based on the correspondence between the word and the business category in the industry-specific terminology 561. In order to accurately represent the contents of the API document by vector generation, the similarity calculation unit 59 gives a higher weight to a word unique to this document that does not appear in other documents. The similarity calculation unit 59 breaks the sentence into words, generates a sentence vector using the number of occurrences for each word as an element, and treats the difference in the size and direction of the vectors related to the two sentences as the similarity between the two sentences. Extract multiple APIs with high similarity. Note that the similarity calculation unit 59 may preferentially select a matching destination that has a similarity greater than or equal to a threshold and the same API name.

  As a result of the calculation by the similarity calculation unit 59, the API recommendation unit 54 presents an API candidate group recommended as a matching destination on the screen (display device) of the user terminal. The API recommendation unit 54 further asks the user to input a correct / incorrect determination result for each API candidate, and stores the correct / incorrect determination result in the correct / incorrect determination result unit 55. The accuracy of the similarity calculation unit 59 is improved by further updating the data in the teacher data holding unit 56 based on the determination result held in the correct / incorrect determination result unit 55. By doing in this way, the calculation accuracy of the similarity calculation unit 59 is improved as the API recommendation is repeated.

The API information holding unit 53 (holding unit) is a database that holds specification information of API_A (first API) related to the service A as a matching destination. The API specification 7 includes a plurality of combinations of a sentence 71 describing each API, a request parameter 72, and a response parameter 73. Note that the API information holding unit 53 may hold API specification information related to a plurality of services as matching destinations. The API information holding unit 53 may hold, for example, API specification information of TMF (TeleManagement FORUM) and API specification information of AWS (Amazon (registered trademark) Web Service).
The teacher data holding unit 56 holds an industry-specific terminology 561. Specifically, it has “a keyword specific to a predetermined task”. For example, let us consider a case where 100 API groups are classified into each classified business category in which business categories are classified into “planning / construction business”, “maintenance business”, and “charging business”. In this case, if an API is recommended regardless of the division information for each business, it is necessary to search from 300 API groups.
However, for example, if the word “Alarm” appears frequently in the API document, assuming that it is an “API to be used for maintenance work”, the candidate is “API group to be used for maintenance work” in recommending the matching destination API. Can be narrowed down to 100. This makes it easier to find the correct API. For such correspondence, for example, keywords may be extracted from API specification information of TMF. Each data of the teacher data holding unit 56 is referred to by the similarity calculation unit 59 and used for weighting the similarity.

The correctness / incorrectness determination result unit 55 holds the correctness / incorrectness determination result by the user. For example, “Template” of cloud ⁿ (registered trademark) is known to be equivalent to “Image” of AWS, but is not regarded as equivalent only by mechanically calculating the similarity. As a result of manual determination, it is preferable to generate and hold a rule that “Template” is equivalent to “Image”.

FIG. 3 is a diagram illustrating an example of the API document 6.
The API document 6 is a document related to the API specification. The API name 61 holds the name of the API, the API description 62 explains the specification related to the API, the request parameter 63, the response parameter 64, the request parameter example 65, and the response parameter example 66. It is comprised including. The request parameter example 65 shows an example of the request parameter 63. In the response parameter example 66, an example of the response parameter 64 is shown.
The API name possessed by the API name 61 is often described in a camel case or snake case that is a compound word. Here, the camel case is a character string in which a space between a plurality of words is filled and a next word is started from an upper case letter. A snake case is a character string in which spaces between a plurality of words are replaced with underscores. This is due to implementation restrictions that only one string token is allowed as an API name.
For example, “alarm name” is described as “AlarmName” when described in a camel case, and “alarm_name” when described as a snake case. Therefore, by decomposing the character string of the camel case or snake case that causes noise in the document vector into each word, the number of words constituting the document vector can be increased and the accuracy of similarity calculation can be improved.
The request parameter example 65 and the response parameter example 66 are often described in JSON format or XML format, or are often described by character strings or words including numbers. Here, the character string or word including a number means, for example, “id = 1234”. Here, the number “1234” is merely an example, and is an unnecessary element for knowing the purpose of use of the API. Therefore, by determining and deleting the JSON format or XML format enclosed in braces {} and angle brackets <> and determining and deleting character strings or words including numbers, request parameter example 65 and response parameter Example 66 can be deleted.

FIG. 4 is a flowchart showing the document processing process and the similarity calculation process, which will be described with reference to FIG.
When the accepting unit 57 accepts the API document 6 input by the user, the process shown in FIG. 4 starts. The accepting unit 57 can accept the two types of text, the entire text in the API document 6 and the text having the request parameter 63 and the response parameter 64. The user may input one API individually or may input a certain API group collectively.

In the API document 6, in addition to the API description 62, the request parameter 63, and the response parameter 64, a request parameter example 65 and a response parameter example 66 are also described. The processing unit 58 deletes the description related to the request parameter example 65 and the response parameter example 66 from the input API document 6 (step S10). Thereby, the noise in the process of the similarity calculation part 59 can be removed.
The similarity calculation unit 59 corrects the text of the API document 6 based on the teacher data holding unit 56 (step S11). More specifically, the similarity calculation unit 59 normalizes the synonyms by replacing the input API document of the service X and the word of the API document group of the service A as a comparison destination. In this way, by utilizing the information of the teacher data holding unit 56, it is possible to absorb the difference in unique expressions that differ for each service and to improve the accuracy of API selection.

  Next, the similarity calculation unit 59 breaks down the text of the API document 6 or API specification 7 into words (step S12), and weights the words unique to the document (step S13). As a result, the contents of the API document can be accurately represented. The similarity calculation unit 59 generates a text vector using the number of occurrences for each word as an element (step S14), and treats the difference in vector size and direction as the similarity.

  If the input API document 6 relates to a plurality of API groups (step S15 → Yes), the similarity calculation unit 59 generates a matching pair that maximizes the overall similarity and positions it as the first candidate ( In step S16), the API group with higher similarity is set as the second candidate to the Nth candidate (step S17), and the candidate group of matching pairs is displayed on a display device (not shown) or the like (step S18). The similarity calculation unit 59 receives input of the correctness / incorrectness determination results of these candidate groups (step S19) and stores them in the correctness / incorrectness determination result unit 55 (step S23).

If the input API document 6 relates to a single API (step S15 → No), the similarity calculation unit 59 sets the API with higher similarity as the first candidate to the Nth candidate (step S20). Are displayed on a display device (not shown) or the like (step S21). The similarity calculation unit 59 accepts input of the correctness determination result of these candidate groups (step S22) and stores it in the correctness determination result unit 55 (step S23). Based on the result stored in the correctness determination result section 55, new teacher data can be generated to improve the calculation accuracy of the similarity calculation section 59.
According to the present embodiment, it is possible to reduce the man-hours for newly creating the API conversion scenario 47 and the man-hours for updating the API conversion scenario 47 accompanying the update of the API specification. Furthermore, new services that do not conform to the common API can be easily entered.

(Modification)
The present invention is not limited to the above-described embodiment, and can be modified without departing from the spirit of the present invention. For example, there are the following (a) and (b).
(A) The processing of the similarity calculation unit 59 is not limited to the flowchart of FIG. For example, the similarity calculation unit 59 may decompose specification information related to API_A to be shared into words in advance, or may be normalized in advance using the industry-specific glossary 561.
(B) The present invention is not limited to sharing the API of the network operator system that uses the existing service. Contrary to the above embodiment, the API of a service provider system that provides an existing service may be converted to a common API and provided to a network provider system that enjoys a service using the common API.

1, 1a, 1b Service provider system 2, 2a, 2b Network provider system 2c Cloud provider system 4 Mediator 45 API converter 47 API conversion scenario 5 Recommendation device (similar document recommendation system)
51 API document input part (input part)
52 API matching section 53 API information holding section (holding section)
54 API recommendation section (Recommendation section)
55 Correctness determination result section 56 Teacher data holding section 561 Industry specific terminology collection 57 Reception section 58 Processing section 59 Similarity calculation section 6 API document 7 API specification

Claims (8)

A holding unit that holds a plurality of API information having sentences describing a plurality of first APIs;
An input unit for receiving an input of an API document having a sentence explaining the second API;
A similarity calculator that calculates the similarity between the API document and the API information;
A recommendation unit that recommends one or more first APIs similar to the second API based on the similarity calculated by the similarity calculation unit;
A similar document recommendation system comprising: The similarity calculation unit corrects a sentence included in the API document using teacher data.
The similar document recommendation system according to claim 1, wherein: The API document has a request parameter example and / or a response parameter example,
The input unit deletes a request parameter example and a response parameter example that the API document has,
The similar document recommendation system according to claim 1, wherein: The API document has a character string or a word including a number,
The input unit deletes a character string or a word including a number that the API document has,
The similar document recommendation system according to claim 1, wherein: The API document includes a camel case or snake case string,
The input unit decomposes the camel case or snake case character string of the API document into words,
The similar document recommendation system according to claim 1, wherein: The recommendation unit stores a user's correct / incorrect determination on the recommendation result of the first API.
The similar document recommendation system according to claim 1, wherein: Holding a plurality of API information having sentences describing a plurality of first APIs;
Receiving an input of an API document having a sentence explaining the second API;
Calculating a similarity between the API document and the API information;
Recommending one or more first APIs similar to the second API based on the similarity;
A method for recommending similar documents, comprising: Holding means for holding a plurality of API information having sentences describing a plurality of first APIs;
An input means for receiving an input of an API document having a sentence explaining the second API;
Similarity calculation means for calculating the similarity between the API document and the API information;
Recommending means for recommending one or more first APIs similar to the second API based on the similarity calculated by the similarity calculating means;
Similar document recommendation program to make computer function as

Images