CN114385688A

CN114385688A - Application program interface API recommendation method and device

Info

Publication number: CN114385688A
Application number: CN202011121916.4A
Authority: CN
Inventors: 苗璐; 白雪珂; 舒南飞; 林文辉
Original assignee: Aisino Corp
Current assignee: Aisino Corp
Priority date: 2020-10-20
Filing date: 2020-10-20
Publication date: 2022-04-22

Abstract

The application relates to the technical field of computers, in particular to an Application Program Interface (API) recommendation method and device, wherein an API set containing candidate APIs is obtained from a preset API database according to retrieval information input by a user; determining a user identifier of a user, and searching historical data of the user according to the user identifier; respectively aiming at each API, calculating the feature similarity between the features of the retrieval information and the features of any one API, and determining the recommendation score of any one API according to the score, the feature similarity, the popularity score and the corresponding weight of any one API; according to the recommendation scores corresponding to the APIs, the APIs meeting the preset recommendation score conditions are generated into API recommendation lists, and the API recommendation lists are displayed to users, so that API recommendation is performed based on multi-dimensional information of historical data, retrieval information and API popularity of the users, and accurate API recommendation can be achieved.

Description

Application program interface API recommendation method and device

Technical Field

The application relates to the technical field of computers, in particular to a recommendation method and device for an Application Program Interface (API).

Background

With the rapid development of information technology, various Application Program Interfaces (APIs) on the internet are increasing, and although rich APIs provide convenience for users, a large number of APIs also bring a difficult problem of how users should select when needing to call the APIs.

In order to solve the problem of selecting the API, in the prior art, the API can be classified, and then based on different API categories, the user is helped to narrow the selection range and further select the required API, but the number of the APIs is increasingly huge as the API is released. The types of the APIs are more and more, and each API category may include a very large number of APIs, so that it is more difficult for a user to select an API, and how to implement accurate recommendation of the API becomes a problem to be solved.

Disclosure of Invention

The embodiment of the application program interface API recommendation method and device is provided to achieve accurate recommendation of API.

The embodiment of the application provides the following specific technical scheme:

an Application Program Interface (API) recommendation method comprises the following steps:

acquiring an API set containing candidate APIs from a preset API database according to retrieval information input by a user, wherein each API corresponds to a popularity score;

determining the user identification of the user, and finding the scores of the user for the APIs according to the user identification;

respectively calculating the feature similarity between the features of the retrieval information and the features of any one API aiming at each API, and determining the recommendation score of any one API according to the score, the feature similarity, the popularity score and the corresponding weight of any one API;

and generating an API recommendation list containing the APIs meeting the preset recommendation score condition according to the recommendation scores corresponding to the APIs, and displaying the API recommendation list to the user.

Optionally, before acquiring the API set including the candidate APIs from the preset API database according to the retrieval information input by the user, the method further includes:

respectively aiming at each user in a preset historical database, acquiring historical record information of any user from the historical database, performing word segmentation processing on the historical retrieval information in the historical record information to obtain a historical retrieval word label, and determining a historical API label according to a historical calling API in the historical record information, wherein the historical record information at least comprises the historical retrieval information and the historical calling API;

and respectively matching the historical search word labels and the historical API labels with labels corresponding to the APIs in the API database aiming at the users, determining the APIs with the highest preset matching values from the APIs in the API database, acquiring popularity scores corresponding to the APIs with the highest preset matching values, taking the APIs with the highest preset matching values which are larger than the threshold value of the popularity scores as the preferred hot APIs of any user, and generating a hot interest API set comprising the preferred hot APIs.

Optionally, the obtaining of the popularity score corresponding to the API with the highest preset matching value specifically includes:

respectively acquiring attribute information of the APIs with the highest matching values, wherein the attribute information at least comprises a historical calling number value and a historical browsing number value of the last time period, a maximum value of the historical calling number value and a maximum value of the historical browsing number value in each time period, and API release time;

and respectively determining the popularity score of any one API according to the historical calling number value and the historical browsing number value of the last period of any one API, the maximum value of the historical calling number value and the maximum value of the historical browsing number value in each time period, a preset cooling coefficient and the API release time aiming at the APIs with the highest preset matching values.

Optionally, determining the user identifier of the user, and before finding the scores of the user for each API according to the user identifier, further include:

performing spectral clustering processing on each user according to the historical API set of each user to obtain various users, wherein the historical API set comprises APIs called by user histories;

respectively aiming at the various users, acquiring a preference API set of each user in any user from the historical database, and generating a scoring matrix according to the corresponding score of each API contained in the preference API set;

and respectively determining the scores of the various users for the various APIs by taking any one score matrix as an input parameter based on the trained score prediction model aiming at each score matrix.

Optionally, obtaining an API set including candidate APIs from a preset API database according to search information input by a user specifically includes:

acquiring a popular interest API set of the user from a historical database according to the user identification of the user;

acquiring a candidate API set from the API database according to the retrieval information input by the user;

and merging the hot interest API set and the candidate API set to generate an API set.

Optionally, obtaining a candidate API set from the API database according to the retrieval information input by the user specifically includes:

acquiring retrieval information input by a user, performing word segmentation processing on the retrieval information to obtain a retrieval word set, and performing word segmentation processing on the retrieval information to obtain a retrieval word set;

acquiring an API (application programming interface) containing at least one search word from a preset API database according to each search word in the search word set;

if the API containing at least one search word is determined to be obtained, generating a candidate API set according to a plurality of preset APIs containing the largest number of search words;

and if determining that the API containing at least one search word is not obtained, obtaining the API containing at least one search word from the API database according to each search word in the search word set, and generating a candidate API set according to a preset API containing the maximum number of search words.

Optionally, if it is determined that the API including at least one search term is obtained, calculating a feature similarity between the feature of the search information and the feature of any API, specifically including:

respectively extracting the features of the search terms, determining the feature vectors of the search terms, and determining the average value of the search vectors of the feature vectors;

acquiring each label word of any one API, and acquiring a label vector evaluation average value of any one API;

and determining the feature similarity between the retrieval information and the any one API according to the retrieval words, the label words, a preset first weight coefficient, the retrieval vector average value and the label vector average value.

Optionally, if it is determined that an API including at least one search term is not obtained, calculating a feature similarity between a feature of the search information and a feature of any API, specifically including:

acquiring each signature of any API;

and determining the feature similarity between the retrieval information and any one API according to the retrieval words, the signature words and a preset second weight coefficient.

An Application Program Interface (API) recommendation device comprising:

the first acquisition module is used for acquiring an API set containing candidate APIs from a preset API database according to retrieval information input by a user, wherein each API corresponds to a popularity score;

the searching module is used for determining the user identification of the user and searching the scores of the user for the APIs according to the user identification;

the first processing module is used for calculating the feature similarity between the features of the retrieval information and the features of any one API aiming at each API, and determining the recommendation score of any one API according to the score, the feature similarity, the popularity score and the corresponding weight of any one API;

and the second processing module is used for generating an API recommendation list containing the APIs meeting the preset recommendation score condition according to the recommendation scores corresponding to the APIs, and displaying the API recommendation list to the user.

the third processing module is used for respectively aiming at each user in a preset historical database, acquiring historical record information of any user from the historical database, performing word segmentation processing on the historical retrieval information in the historical record information to obtain a historical retrieval word label, and determining a historical API label according to a historical calling API in the historical record information, wherein the historical record information at least comprises the historical retrieval information and the historical calling API;

and the matching module is used for respectively matching the historical search word labels and the historical API labels with labels corresponding to the APIs in the API database aiming at the users, determining the APIs with the highest preset matching values from the APIs in the API database, acquiring popularity scores corresponding to the APIs with the highest preset matching values, taking the APIs with the highest preset matching values which are larger than the threshold value of the popularity scores as the preferred hot APIs of any user, and generating a hot interest API set comprising the preferred hot APIs.

Optionally, when obtaining the popularity score corresponding to the API with the highest matching value of the preset number, the matching module is specifically configured to:

the obtaining module is used for carrying out spectral clustering processing on each user according to the historical API set of each user to obtain various users, wherein the historical API set comprises the API called by the user history;

a second obtaining module, configured to obtain, for each of the types of users, a preference API set of each user in any type of users from the historical database, and generate a score matrix according to a score corresponding to each API included in the preference API set;

and the determining module is used for determining the scores of the various users for the various APIs based on the trained score prediction model and any one score matrix as an input parameter respectively aiming at the various score matrixes.

Optionally, the first obtaining module is specifically configured to:

Optionally, when the candidate API set is obtained from the API database according to the retrieval information input by the user, the first obtaining module is specifically configured to:

Optionally, if it is determined that an API including at least one search term is obtained, when the feature similarity between the feature of the search information and the feature of any API is calculated, the first processing module is specifically configured to:

Optionally, if it is determined that an API including at least one search term is not obtained, when the feature similarity between the feature of the search information and the feature of any API is calculated, the first processing module is specifically configured to:

acquiring each signature of any API;

An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the API recommendation method when executing the program.

A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the API recommendation method described above.

In the embodiment of the application, an API set containing candidate APIs is obtained from a preset API database according to retrieval information input by a user, a user identification of the user is determined, scores of the user for the APIs are found according to the user identification, feature similarity between the features of the retrieval information and the features of any one API is calculated respectively for the APIs, and the recommendation score of any one API is determined according to the score, the feature similarity, the popularity score and the corresponding weight of any one API, so that the recommendation score of each API can be determined, an API recommendation list containing the APIs meeting the preset recommendation score condition is generated according to the recommendation score corresponding to each API, and the API recommendation list is displayed to the user, so that historical data of the user is obtained, and multi-dimensional information such as the scores of the APIs, the popularity scores of the APIs and the feature similarity is based on the API recommendation list, the recommendation score of the API is calculated, accurate API recommendation can be achieved, the recommendation score of the API is calculated based on historical data of a user, deep mining of user requirements can be achieved, the API is recommended according to the user requirements, and accuracy of API recommendation is further guaranteed.

Drawings

FIG. 1 is a flowchart of an API recommendation method in an embodiment of the present application;

FIG. 2 is a technical roadmap for API recommendations in an embodiment of the present application;

FIG. 3 is a schematic structural diagram of an API recommendation device in an embodiment of the present application;

fig. 4 is a schematic structural diagram of an electronic device in an embodiment of the present application.

Detailed Description

The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

In order to solve the problem of selecting the API, in the prior art, the API may be classified, for example, the categories such as speech technology, text recognition, face recognition, natural language processing, and the like, and based on these classifications, the API may be further subdivided to help the user narrow the selection range and further accurately select the required API.

When the user defines the description of the API required by the user and the number of similar APIs is small, the manner of searching for the API classification and direct searching can screen out the relevant APIs for further selection for the user, but the number of the APIs is increasingly huge as the API is published. The types of the APIs are more and more, each API category may include a very large number of APIs, so that it is more difficult for a user to select the APIs, and the retrieved APIs are provided by publishers in the network, and sometimes the descriptions of the user's requirements for the APIs are blurred, so how to implement accurate recommendation of the APIs becomes a problem to be solved.

In the embodiment of the application, an API set containing candidate APIs is obtained from a preset API database according to retrieval information input by a user, a user identification of the user is determined, scores of the user for the APIs are found according to the user identification, feature similarity between the features of the retrieval information and the features of any one API is calculated respectively aiming at the APIs, and the recommendation score of any one API is determined according to the score, the feature similarity, the popularity score and the corresponding weight of any one API, so that the recommendation score of each API can be determined, an API recommendation list containing the APIs meeting the preset recommendation score condition is generated according to the recommendation score corresponding to each API, and the API recommendation list is displayed to the user, so that the three-angle comprehensive evaluation APIs of the recommendation score of each API are calculated from historical data of the user, the API and the score and the feature similarity, the API recommendation method and the API recommendation device can realize accurate recommendation of the API, recommend the API for the user based on the historical data of the user, can realize deep mining of user requirements, and guarantee accuracy of API recommendation.

Based on the foregoing embodiment, referring to fig. 1, a flowchart of an API recommendation method in an embodiment of the present application is specifically included:

step 100: and acquiring an API set containing candidate APIs from a preset API database according to retrieval information input by a user.

Wherein each API corresponds to a popularity score.

In the embodiment of the application, the retrieval information input by the user is obtained in real time, the API relevant to the retrieval information is searched in the preset API database according to the retrieval information input by the user, and the searched API is generated into the API set, so that the API set containing the candidate APIs can be obtained from the API database according to the retrieval information input by the user, and when the API set containing the candidate APIs is obtained, the popularity scores corresponding to the APIs are obtained at the same time, and each API corresponds to one popularity score.

Specifically, when step 100 is executed, the method specifically includes:

s1: and acquiring the popular interest API set of the user from the historical database according to the user identification of the user.

In the embodiment of the application, when the retrieval information input by the user is acquired, the user identification of the user can be acquired at the same time, and then the popular interest API set corresponding to the user identification can be acquired from the historical database according to the user identification of the user.

The method for acquiring the user identifier of the user may be, for example, that the user can perform the retrieval only by performing a login operation before retrieving the required API, so that the user identifier of the user can be acquired when the user performs the login operation, and the method for acquiring the user identifier of the user is not limited in the embodiment of the present application.

The following is a detailed description of the steps of generating the popular interest API set in the embodiment of the present application, and specifically includes:

a1: the method comprises the steps of respectively aiming at each user in a preset historical database, obtaining historical record information of any user from the historical database, carrying out word segmentation processing on historical retrieval information in the historical record information to obtain a historical retrieval word label, calling an API according to the history in the historical record information, and determining the historical API label.

The history information at least comprises history retrieval information and a history calling API.

In the embodiment of the present application, the following operations are performed for each user in each preset history database, respectively:

the method comprises the steps of obtaining historical record information of any user from a historical database, wherein the historical record information at least comprises historical retrieval information, historical API (application program interface) calling information and historical browsing record information, and then deleting repeated data in the historical retrieval information and the historical API calling information.

The historical retrieval information and the attribute information called by the historical API comprise attribute information, the attribute information at least comprises data state information, the data state information of the historical retrieval information represents whether the historical retrieval information is missing or abnormal data, the data state information called by the historical API represents whether the historical retrieval information called by the API is missing or abnormal data, then, according to the data state information of the historical retrieval information and the data state information called by the historical API, the historical retrieval information with the data state information being abnormal or missing is deleted, and the historical API calling information with the data state information being abnormal or missing is deleted.

And because the historical retrieval information is mainly Chinese and English character strings, the historical retrieval information is subjected to word segmentation processing, each historical retrieval word of the historical retrieval information is determined, each historical retrieval word label of the historical retrieval information is further obtained, meanwhile, the historical API labels corresponding to each historical called API in the historical API calling information are obtained, word segmentation processing is carried out on the labels, a stop word list is maintained, and stop words, punctuation marks and special marks are removed.

Therefore, the historical search terms and the historical API labels corresponding to each user in the historical database can be obtained, and the historical search term labels and the historical API labels jointly form key features reflecting user preferences.

The historical retrieval information is retrieval information input by the user history, and the historical API calling information is an API called by the user history.

It should be noted that the historical API tags in the historical database are tags that are pre-marked for each API, each API in the API database has a chinese or english tag, and word segmentation processing and maintenance of the stop word list are also required to be performed, and tag features are extracted as the tags of the API.

Further, when extracting the history information of any user from the history database, the operation time may be acquired, that is, when acquiring the history search information in the history information, the search time may be acquired, when acquiring the history API call information, the API call time may be acquired, and when acquiring the history browsing information, the browsing API may be acquired.

A2: and respectively matching the historical search word labels and the historical API labels with labels corresponding to all APIs in an API database aiming at all users, determining the APIs with the highest preset matching values from all the APIs in the API database, acquiring popularity values corresponding to the APIs with the highest preset matching values, taking the APIs with the highest preset matching values which are larger than the threshold value of the popularity values as hot APIs preferred by any user, and generating a hot interest API set comprising the hot APIs with all the preferences.

In the embodiment of the application, for each user in the history database, firstly, a user tag set is generated according to a history search word tag of any user history search information and a history API tag of a history call API, and the user tag set includes the history search word tag and the history API tag of the user.

And then, matching the user tag set with the API tag sets corresponding to the APIs in the API database to obtain the APIs corresponding to the API tag sets with intersection with the user tag set, determining matching values between the API tag sets and the APIs, determining a plurality of APIs with the highest preset matching values from the searched APIs, and generating the user interest API set according to the searched APIs.

Then, popularity scores corresponding to the APIs in the user interest API set are obtained, whether the popularity score corresponding to the API is larger than a preset popularity score threshold value or not is judged for each API, the API larger than the preset popularity score threshold value is used as a hot API preferred by any user, and the determined hot API preferred by each user is used for generating a hot interest API set of any user.

The following describes the obtaining step of the popularity score of each API in the embodiment of the present application, and specifically includes:

n1: and respectively acquiring attribute information of a plurality of preset APIs with the highest matching values.

The attribute information at least comprises a historical calling number value and a historical browsing number value of the last time period, a maximum value of the historical calling number value and a maximum value of the historical browsing number value in each time period, and API release time.

In the embodiment of the application, a plurality of preset APIs with the highest matching values are determined from the APIs in the API database, and meanwhile attribute information of the APIs obtained through matching is obtained.

N2: and respectively determining the popularity score of any API according to the historical calling number value and the historical browsing number value of the last period, the maximum value of the historical calling number value and the maximum value of the historical browsing number value in each time period, and the preset cooling coefficient and the API release time aiming at a plurality of preset APIs with the highest matching values.

In the embodiment of the application, the period duration is preset for each API obtained through matching, and the popularity score of any one API is calculated and obtained according to the historical calling number value of the last period, the historical browsing number value of the last period, the maximum value of the historical calling number value, the maximum value of the historical browsing number value, the preset cooling coefficient and the API release time.

Specifically, the popularity score in the embodiment of the present application may be expressed as:

wherein, a is the historical calling times value of the API in the last time period, b is the historical browsing times value of the API in the last time period, a_maxFor the maximum value of the number of calls in the history, i.e. the maximum value of the number of calls in the history during the respective time period, b_maxThe maximum value of the browsing times in the history record, namely the maximum value of the historical browsing times in each time period, k is a preset cooling coefficient, k is a coefficient and represents the cooling speed, and h represents the time from the moment when the API is issued.

It should be noted that the maximum value of the historical recall count value and the maximum value of the historical browse count value in the embodiment of the present application are constant values in each round of calculation, and in order to prevent a situation that a certain API is browsed to cause an excessive popularity of the API for a long time due to a short-time explosive recall, a natural cooling model, that is, a preset cooling coefficient, is introduced.

S2: and acquiring a candidate API set from the API database according to the retrieval information input by the user.

In the embodiment of the application, the candidate API set is obtained from the API database according to the retrieval information input by the user.

Specifically, when step S2 is executed, the method specifically includes:

a1: the method comprises the steps of obtaining retrieval information input by a user, carrying out word segmentation processing on the retrieval information to obtain a retrieval word set, and carrying out word segmentation processing on the retrieval information to obtain a retrieval word set.

In the embodiment of the application, after the retrieval information input by the user is acquired, word segmentation processing is performed on the retrieval information to acquire each retrieval word included in the retrieval information and generate a retrieval word set including each retrieval word, and word segmentation processing is performed on the retrieval information to acquire each retrieval word included in the retrieval information and generate a retrieval word set including each retrieval word.

A2: and acquiring an API (application programming interface) containing at least one search term from a preset API database according to each search term in the search term set.

In the embodiment of the application, according to each search term in the search term set, searching is performed in a preset API database, whether an API tag corresponding to each API in the API database contains at least one search term is judged, and if it is determined that the API tag corresponding to one API contains at least one search term, the API is obtained.

A3: and if the API containing at least one search word is determined to be obtained, generating a candidate API set according to a plurality of preset APIs containing the largest number of search words.

In the embodiment of the application, whether the API containing at least one search word can be acquired is judged, if the API containing at least one search word is determined to be acquired, a certain API or certain APIs can be retrieved according to the search word in the search information, and a plurality of preset APIs containing the largest number of search words are generated to form a candidate API set.

A4: and if determining that the API containing at least one search word is not obtained, obtaining the API containing at least one search word from the API database according to each search word in the search word set, and generating a candidate API set according to a preset API containing the maximum number of search words.

In the embodiment of the application, whether an API containing at least one search word can be acquired is judged, if it is determined that the API containing at least one search word cannot be acquired, that is, the search word in the search information is not matched with a tag corresponding to the API, whether the API containing at least one search word can be acquired is judged according to each search word in the search word set, and if it is determined that the API containing at least one search word can be acquired, a plurality of preset APIs with the largest number of search words are included to generate a candidate API set.

S3: and merging the preference API set and the candidate API set to generate an API set.

In the embodiment of the application, after the preference API set of the user is obtained according to the user identification of the user and the candidate API set is obtained from the API database, the preference API set and the candidate API set are merged to generate the API set.

Further, in order to reduce the amount of computation, duplicate APIs may also be deleted when merging the preference API set and the candidate API set.

Step 110: and determining the user identification of the user, and searching the historical data of the user according to the user identification.

Wherein the historical data at least comprises the scores of the APIs.

In the embodiment of the application, the user identification of the user is determined, the historical data of the user corresponding to the user identification is searched according to the user identification, and the historical data at least comprises the scores of the APIs, so that the scores of the user for the APIs in the API set can be obtained.

Specifically, when history information of any user is extracted from the history database, the method specifically includes:

n1: and performing spectral clustering processing on each user according to the historical API set of each user to obtain various users.

And the history API set comprises the API called by the user history.

In the embodiment of the application, the historical API set of each user is obtained from the historical database, the historical API set comprises the API which is called by each user in a historical mode, and then, each user is clustered based on a spectral clustering technology to obtain various users.

Specifically, when clustering is performed, for an undirected graph G (V, E), V is a point in the graph, the point represents a user, E is an edge in the graph, an edge between two points represents that two users have called the same API, a weight value of the edge represents the number of the same API called by the two users, and when two users have not called the same API, no edge exists between two points, that is, the weight value is 0. The adjacency matrix W is a matrix composed of weight values W _ ij between any two points, the matrix is a symmetric matrix, and the degree matrix D is calculated by the following formula:

that is, the sum D of each row element of the similarity matrix W is D_iForming an n x n diagonal matrix. Calculating a Laplace matrix L-D-W, calculating eigenvalues of L, sorting the eigenvalues from small to large, taking the first k eigenvalues, calculating eigenvectors of the first k eigenvalues, and forming a matrix U-U by the k column vectors₁,u₂,..,u_k},U∈R^n*kLet y_i∈R^kIs the vector of the ith row of U, where i is 1,2, …, n, and the new sample point Y is { Y ═ Y using the k-means algorithm₁,y₂,..,y_nCluster-forming C₁,C₂,..,C_kOutput cluster A₁,A₂,..,A_kWherein A is_i＝{j|y_j∈C_iAnd spectral clustering has stronger adaptability to data distribution, good clustering effect and smaller computation amount of clustering, and is suitable for clustering users.

N2: and respectively aiming at various users, acquiring a preference API set of each user in any user from the historical database, and generating a scoring matrix according to the corresponding score of each API contained in the preference API set.

In the embodiment of the application, the following operations are executed respectively for various users:

according to the user identification corresponding to each user in any type of users, acquiring a preference API set corresponding to each user identification from a historical database, wherein each user identification corresponds to one preference API set, each preference API set comprises at least one API and the corresponding scores of the API, so that a scoring matrix is generated according to the corresponding scores of the API in the preference API set, each row of the scoring matrix is each user, and each column is the score of each user for the API.

If the user does not call any API, the corresponding score is 0.

Specifically, after users are grouped, the users in each group respectively search the corresponding preference API set, user-API scoring based on matrix decomposition is performed, user preferences are mined, historical evaluation data of the users on the API can be represented as a user-API scoring matrix, the matrix is a sparse matrix, the users call certain API service for 5 points, browsing API for 1 point and the rest are empty, the scores r of the unknown users on the API are predicted by using the existing part of sparse data, and the loss function of matrix decomposition is as follows:

wherein, I_ijTo indicate a function, when a user has a call or a browse operation to a service, i.e. r_ijNot being empty, I_ijBy continuously optimizing the scoring matrix, the score of the user on the unknown API can be predicted, and the score r of the user for each API service can be obtained.

N3: and respectively determining the scores of various users for various APIs (application program interfaces) based on the trained score prediction model and any one score matrix as an input parameter aiming at each score matrix.

In the embodiment of the application, after the scoring matrices are obtained, any one scoring matrix is input into a trained scoring prediction model for each scoring matrix, and the scoring of each user of any type for each API is determined.

It should be noted that, with the real-time update of the data, the score prediction model and parameters need to be updated and maintained regularly, the change trend of the user interest and popularity is mined, the real-time performance of recommendation is ensured, and in the training process, a large number of parallel repeated calculations are performed by using the GPU, so that the training speed of the score prediction model can be improved effectively.

Therefore, the users are grouped by adopting the spectral clustering technology, the user-API scores are respectively calculated for the users in different groups, the model is updated more flexibly, the score of the users is predicted by taking the class as a basic unit, the calculation difficulty of mass data can be reduced, the user-API scores are calculated off line based on a matrix decomposition method after the users are grouped, and the API retrieval recommendation speed and accuracy can be improved.

Step 120: and respectively calculating the feature similarity between the features of the retrieval information and the features of any one API aiming at each API, and determining the recommendation score of any one API according to the score, the feature similarity, the popularity score and the corresponding weight of any one API.

In the embodiment of the present application, before calculating the recommendation score of any one API for each API, first, a similarity between the retrieval information and any one API needs to be obtained, and the following steps of calculating the similarity between the retrieval information and any one API in the embodiment of the present application are described in detail, and may be specifically divided into the following two ways according to the retrieval information.

The first mode specifically includes:

s1: and respectively extracting the features of the search words, determining the feature vectors of the search words, and determining the average value of the search vectors of the feature vectors.

In the embodiment of the application, based on a trained feature extraction model, feature extraction is performed on each search term by taking each search term as an input parameter, feature vectors of each search term are determined, and then after the feature vectors of each search term are obtained, the feature vectors of each search term are averaged, so that a search vector average value among the feature vectors is obtained.

S2: and acquiring each label word of any one API, and acquiring the average value of the label vectors of any one API.

In the embodiment of the application, any one API label word is obtained, feature extraction is performed on each label word by taking each label word as an input parameter based on a trained feature extraction model, feature vectors of each label word are determined, and then after the feature vectors of each label word are obtained, the feature vectors of each label word are averaged to obtain a label vector average value among the feature vectors.

For example, feature extraction may be performed based on the word2vec model to obtain a feature vector corresponding to the tag word.

S3: and determining the feature similarity between the retrieval information and any API according to each retrieval word, each label word, a preset first weight coefficient, a retrieval vector average value and a label vector average value.

In the embodiment of the application, a search term set is obtained according to each search term, a label term set is obtained according to each label term, then a union set between the search term set and the label term set is calculated, and an intersection set between the search term set and the label term set is calculated.

After the union set and the intersection set are determined, determining the feature similarity between the retrieval information and any one API according to the union set, the intersection set, the preset first weight coefficient, the retrieval vector average value and the tag vector average value, where the feature similarity between the retrieval information and any one API may be expressed as:

wherein S is₁For a set of search terms, A₁As a set of tagged words, mu₁Is a preset first weight coefficient, w_sTo retrieve the vector mean, w_aIs the tag vector average.

It should be noted that, in the following description,

the intersection of the search word and all the API label word sets is not null, namely the search word can be matched with a certain API or certain APIs, similarity is calculated according to word matching, and when the search word is not matched with the API, factors such as typing errors of a user are considered, and word matching is carried out.

The second mode specifically includes:

s1: tag signals of any one API are obtained.

In the embodiment of the application, a signature of any one API is obtained.

S2: and determining the feature similarity between the retrieval information and any API according to each retrieval word, each signature word and a preset second weight coefficient.

In the embodiment of the application, a search word set is obtained according to each search word, a tag word set is obtained according to each tag word, then a union set between the search word set and the tag word set is calculated, and an intersection set between the search word set and the tag word set is calculated.

After the union set and the intersection set are determined, determining the feature similarity between the retrieval information and any one API according to the union set, the intersection set and a preset second weight coefficient, where the feature similarity between the retrieval information and any one API may be expressed as:

wherein S is₂For search word sets, A₂Is a set of label words, mu₂Is a preset second weight coefficient.

And after the feature similarity between the retrieval information and any one API is determined, determining the recommendation score of any one API according to the feature similarity, the preset weight and the popularity score.

The recommendation score for an API may be expressed, for example, as:

S＝sim+λ₁·r+λ₂·pop

wherein λ is₁And λ₂Different weights are given to the total scores, and the weight values are determined through historical data.

Step 130: and generating an API recommendation list containing the APIs meeting the preset recommendation score condition according to the recommendation scores corresponding to the APIs, and displaying the API recommendation list to the user.

In the embodiment of the application, after the recommendation score corresponding to each API is determined, the API meeting the preset recommendation score condition generates an API recommendation list, and the API recommendation list is displayed to a user.

In the embodiment of the application, an API set containing candidate APIs is obtained from a preset API database according to retrieval information input by a user, a user identification of the user is determined, the user scores of the API are found according to the user identification, the feature similarity between the features of the retrieval information and the features of any one API is calculated respectively aiming at the APIs, the recommendation score of any one API is determined according to the score, the feature similarity, the popularity score and the corresponding weight of any one API, thus the recommendation score of each API can be determined, an API recommendation list containing the APIs meeting the preset recommendation score condition is generated according to the recommendation score corresponding to each API, and the API recommendation list is displayed to the user, so that when the APIs are retrieved from the API database according to the retrieval information, retrieval is realized based on the historical data of the user, the retrieval information and the popularity of the APIs, therefore, on the basis of ensuring the rapid and accurate recommendation, the novelty of the API recommendation is increased, in the recommendation process, the search information is mainly retrieved, the search content is ensured not to deviate from the search subject, on the basis, the interest and popularity of the user are combined, comprehensive recommendation is carried out from multiple dimensions, and therefore accurate recommendation of the API can be achieved.

Based on the foregoing embodiment, referring to fig. 2, a technical route diagram recommended by an API in the embodiment of the present application specifically includes:

1. user history data.

In the embodiment of the application, the history information of each user is obtained from the history database, and the history information at least comprises history retrieval information and a history calling API, and also comprises history API browsing record information.

2. And (5) processing historical data.

In the embodiment of the application, repeated data is deleted, and missing and abnormal data are processed.

And performing word segmentation processing on the historical search information to obtain a historical search word tag, calling an API (application programming interface) according to the history in the historical record information, and determining the historical API tag, wherein the historical search word tag and the historical API tag jointly form a key feature which reflects the interest of the user.

3. The popularity score of the API is calculated.

In the embodiment of the application, the popularity score of each API is determined according to the historical calling number and the historical browsing number of the API in the last period, the maximum value of the historical calling number and the maximum value of the historical browsing number in each time period, and the preset cooling coefficient and the API release time.

It should be noted that, in the embodiment of the present application, when the popularity score of the API is calculated, a cooling coefficient is introduced, so that a situation that the popularity of a certain API is too high for a long time due to a short-time explosive call and browsing of the API can be effectively prevented.

4. The API is recalled offline.

In the embodiment of the application, matching is performed in each API according to the historical search term tags and the historical API tags, an API set which is interested by a user is selected in an off-line mode, APIs with matching tags are reserved, and the APIs are updated regularly.

And according to the popularity scores corresponding to the APIs, taking the APIs which are larger than a preset popularity score threshold value as hot APIs, and selecting the hot API set from the API sets which are interested by the user to store as the hot interest API set.

5. The information is retrieved.

In the embodiment of the application, the retrieval information of the user is acquired.

6. An API tag.

In the embodiment of the application, a plurality of APIs are stored in a preset API database, and each API corresponds to at least one API tag.

7. And (5) characterizing.

In the embodiment of the application, the features comprise character features and word features, the character features comprise search words and label words, the word features comprise search words and label words, feature extraction is performed on the search words respectively, feature vectors of the search words are determined, feature extraction is performed on the label words respectively, and feature vectors of the label words are obtained.

8. Feature similarity.

In the embodiment of the application, the API service containing the search word and the search word is searched in the API library, the first N APIs with the largest number of matching words are selected from the massive APIs, and the special similarity calculation is carried out on the first N APIs and the popular interest API set of the user.

9. user-API scoring.

In the embodiment of the application, spectral clustering processing is carried out on each user according to the historical API set of each user to obtain various users, and the score of the unknown user to the API is predicted based on a matrix decomposition technology.

10. A score is recommended.

In the embodiment of the application, the recommendation score of the API is determined according to the score, the feature similarity, the popularity score and the corresponding weight of the API.

In the embodiment of the application, multi-dimensional information such as historical data, retrieval information and API popularity of a user is considered, user requirements are deeply mined, accurate user recommendation is achieved, user clustering is carried out by adopting a spectral clustering method, user-API scores are respectively calculated for users in different groups, the model is more flexible to update, and the calculation difficulty of mass data is reduced.

Based on the same inventive concept, the embodiment of the present application further provides an API recommendation device, which may be a hardware structure, a software module, or a hardware structure plus a software module. Based on the above embodiments, referring to fig. 3, a schematic structural diagram of an API recommendation device in the embodiment of the present application is shown, which specifically includes:

a first obtaining module 300, configured to obtain, according to search information input by a user, an API set including candidate APIs from a preset API database, where each API corresponds to a popularity score;

the searching module 310 is configured to determine a user identifier of the user, and search for a score of each API for the user according to the user identifier;

a first processing module 320, configured to calculate, for each API, a feature similarity between a feature of the search information and a feature of any one API, and determine a recommendation score of the any one API according to a score, the feature similarity, a popularity score, and a corresponding weight of the any one API;

the second processing module 330 is configured to generate an API recommendation list including APIs meeting a preset recommendation score condition according to the recommendation scores corresponding to the APIs, and display the API recommendation list to the user.

a third processing module 340, configured to obtain, for each user in a preset history database, history record information of any user from the history database, perform word segmentation processing on the history search information in the history record information, obtain a history search word tag, and determine a history API tag according to a history call API in the history record information, where the history record information at least includes history search information and a history call API;

a matching module 350, configured to match, for each user, the historical search term tag and the historical API tag with a tag corresponding to each API in the API database, determine, from the APIs in the API database, an API with a preset number of highest matching values, obtain a popularity score corresponding to the API with the preset number of highest matching values, use, as a hot API preferred by the arbitrary user, the API larger than a preset popularity score threshold, and generate a hot interest API set including the hot APIs with the preferences.

Optionally, when obtaining the popularity scores corresponding to the APIs with the highest matching values, the matching module 350 is specifically configured to:

an obtaining module 360, configured to perform spectral clustering processing on each user according to a historical API set of each user to obtain each type of user, where the historical API set includes an API called by a user history;

a second obtaining module 370, configured to obtain, for each of the types of users, a preference API set of each user in any type of users from the history database, and generate a score matrix according to a score corresponding to each API included in the preference API set;

and the determining module 380 is configured to determine, for each scoring matrix, the score of each user for each API based on the trained scoring prediction model and using any one scoring matrix as an input parameter.

Optionally, the first obtaining module 300 is specifically configured to:

Optionally, when the candidate API set is obtained from the API database according to the retrieval information input by the user, the first obtaining module 300 is specifically configured to:

Optionally, if it is determined that an API including at least one search term is obtained, when the feature similarity between the feature of the search information and the feature of any API is calculated, the first processing module 320 is specifically configured to:

Optionally, if it is determined that an API including at least one search term is not obtained, when the feature similarity between the feature of the search information and the feature of any API is calculated, the first processing module 320 is specifically configured to:

acquiring each signature of any API;

Based on the above embodiments, referring to fig. 4, a schematic structural diagram of an electronic device in an embodiment of the present application is shown.

An embodiment of the present application provides an electronic device, which may include a processor 410 (CPU), a memory 420, an input device 430, an output device 440, and the like, wherein the input device 430 may include a keyboard, a mouse, a touch screen, and the like, and the output device 440 may include a Display device, such as a Liquid Crystal Display (LCD), a Cathode Ray Tube (CRT), and the like.

Memory 420 may include Read Only Memory (ROM) and Random Access Memory (RAM), and provides processor 410 with program instructions and data stored in memory 420. In the embodiment of the present application, the memory 420 may be used to store a program of any one of the API recommendation methods in the embodiment of the present application.

The processor 410 is configured to execute any API recommendation method in the embodiments of the present application according to the obtained program instructions by calling the program instructions stored in the memory 420.

Based on the above embodiments, in the embodiments of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program, when executed by a processor, implements the API recommendation method in any of the above method embodiments.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

It will be apparent to those skilled in the art that various changes and modifications may be made in the present application without departing from the spirit and scope of the application. Thus, if such modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such modifications and variations as well.

Claims

1. An API recommendation method, comprising:

2. The method of claim 1, wherein before obtaining the API set including the candidate APIs from the predetermined API database according to the search information input by the user, the method further comprises:

3. The method of claim 2, wherein obtaining the popularity scores corresponding to the APIs with the highest matching values comprises:

4. The method of claim 2, wherein determining the user identifier of the user and finding the scores of the user for the APIs according to the user identifier further comprises:

5. The method of claim 1, wherein obtaining an API set including candidate APIs from a predetermined API database according to search information input by a user comprises:

6. The method of claim 5, wherein obtaining a candidate API set from the API database according to the search information input by the user specifically comprises:

7. The method according to claim 6, wherein if it is determined that an API including at least one search term is obtained, calculating a feature similarity between the feature of the search information and the feature of any one API, specifically includes:

8. The method according to claim 6, wherein if it is determined that an API including at least one search term is not obtained, calculating a feature similarity between the feature of the search information and the feature of any one API, specifically includes:

acquiring each signature of any API;

9. An Application Program Interface (API) recommendation apparatus, comprising:

the acquisition module is used for acquiring an API set containing candidate APIs from a preset API database according to retrieval information input by a user, wherein each API corresponds to a popularity score;

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the steps of the method of any of claims 1-8 are implemented when the program is executed by the processor.

11. A computer-readable storage medium having stored thereon a computer program, characterized in that: the computer program, when being executed by a processor, realizes the steps of the method of any one of claims 1 to 8.