Feature extraction system and method for enhancing recommendations
A method, performed by at least one processor of a computing device, for recommending an event, the method including determining a frequency of execution of the event by a user among a plurality of users; determining an inverse user frequency of execution of the event based on the plurality of users who have executed the event; generating a recommendation factor by modifying the frequency of execution of the event by the user based on the inverse user frequency of execution of the event; and outputting a recommendation for the event based on the recommendation factor.
1. A method, performed by at least one processor of a computing device, for recommending an event, the method comprising:
receiving a request for a recommendation of an event to a target user;
retrieving data from a storage device, wherein the data indicates a number of executions of each of a plurality of candidate events by each of a plurality of users including the target user;
determining, based on the retrieved data, a first metric of the target user for each of the candidate events, wherein the first metric of the target user is calculated based on a ratio of a number of executions of one of the candidate events by the target user to a total number of executions of the candidate events by the target user;
determining, based on the retrieved data, a second metric for each of the candidate events, wherein the second metric is calculated based on a ratio of a total number of the plurality of users to a number of users who have executed one of the candidate events;
generating and storing in the storage device a recommendation factor for each of the candidate events by modifying storing in the storage device the first metric for each of the candidate events based on the second metric for each of the candidate events,
wherein the recommendation factor is modified according to:
Recommendation Factor ij =t ij /Σ m k=1 t ik ×log a |U|/|{uεU:x j εu}|, where t ij /Σ m k=1 t ik represents an execution frequency ij and log a |U|/|{uεU:x j εu}|represents an inverse user frequency i , where i is an integer from 1 to n and n=a number of users U, where j is an integer from 1 to m and m=a number of events X, and where log a is a logarithm base a where a is a real number, wherein the first metric includes the execution frequency ij , wherein the second metric includes the inverse user frequency j ;
calculating, and storing in the storage device based on a machine learning algorithm, a result using the recommendation factor for each of the candidate events as input;
selecting and storing in the storage device an event among the plurality of candidate events based on the result of the calculating; and
presenting on a display the selected event.
2. The method according to claim 1 , further comprising inputting the recommendation factor into a feature vector for input to a recommendation algorithm that determines a possibility of the user executing the event.
3. The method according to claim 2 , wherein the recommendation algorithm is one or more of a logistic regression algorithm, a support vector machine (SVM) algorithm, and an Extreme Gradient Boosting (XGBoost) algorithm.
4. An apparatus for generating a recommendation, the apparatus comprising:
a data storage configured to store an event execution record comprising information concerning a plurality of events executed by a plurality of users; and
at least one computer readable memory having recorded therein program instructions;
at least one processor configured to execute the program instructions, that when executed comprise:
receive a request for a recommendation of an event to a target user;
retrieve data from a storage device, wherein the data indicates a number of executions of each of a plurality of candidate events by each of a plurality of users including the target user;
determine, based on the retrieved data, a first metric of the target user for each of the candidate events, wherein the first metric of the target user is calculated based on a ratio of a number of executions of one of the candidate events by the target user to a total number of executions of the candidate events by the target user;
determine, based on the retrieved data, a second metric for each of the candidate events, wherein the second metric is calculated based on a ratio of a total number of the plurality of users to a number of users who have executed one of the candidate events;
generate and store in the storage device a recommendation factor for each of the candidate events by modifying and storing in the storage device the first metric for each of the candidate events based on the second metric for each of the candidate events,
wherein the recommendation factor is modified according to:
Recommendation Factor ij =t ij /Σ m k=1 t ik ×log a |U|/|{uεU:x j εu}|, where t ij /Σ m k=1 t ik represents an execution frequency ij and log a |U|/|{uεU:x j εu}|represents an inverse user frequency i , where i is an integer from 1 to n and n=a number of users U, where i is an integer from 1 to m and m=a number of events X, and where log a is a logarithm base a where a is a real number, wherein the first metric includes the execution frequency ij , wherein the second metric includes the inverse user frequency j ;
calculate, and store in the storage device based on a machine learning algorithm, a result using the recommendation factor for each of the candidate events as input;
select and store in the storage device an event among the plurality of candidate events based on the result of the calculating; and
present on a display the selected event.
5. The apparatus according to claim 4 , wherein the recommendation factor is input into a feature vector for input to a recommendation algorithm that determines a possibility of the user executing the event.
6. The apparatus according to claim 5 , wherein the recommendation algorithm is one or more of a logistic regression algorithm, a support vector machine (SVM) algorithm, and an Extreme Gradient Boosting (XGBoost) algorithm.
7. A non-transitory computer-readable medium comprising program instructions, that when executed by one or more processors, performs a method comprising:
receiving a request for a recommendation of an event to a target user;
retrieving data from a storage device, wherein the data indicates a number of executions of each of a plurality of candidate events by each of a plurality of users including the target user;
determining, based on the retrieved data, a first metric of the target user for each of the candidate events, wherein the first metric of the target user is calculated based on a ratio of a number of executions of one of the candidate events by the target user to a total number of executions of the candidate events by the target user;
determining, based on the retrieved data, a second metric for each of the candidate events, wherein the second metric is calculated based on a ratio of a total number of the plurality of users to a number of users who have executed one of the candidate events;
generating and storing in the storage device a recommendation factor for each of the candidate events by modifying and storing in the storage device the first metric for each of the candidate events based on the second metric for each of the candidate events,
wherein the recommendation factor is modified according to:
Recommendation Factor ij =t ij /Σ m k=1 t ik ×log a |U|/|{uεU:x j εu}|, where t ij /Σ m k=1 t ik represents an execution frequency ij and log a |U|/|{uεU:x j εu}|represents an inverse user frequency i , where i is an integer from 1 to n and n=a number of users U, where i is an integer from 1 to m and m=a number of events X, and where log a is a logarithm base a where a is a real number, wherein the first metric includes the execution frequency ij , wherein the second metric includes the inverse user frequency j ,
calculating, and storing in the storage device based on a machine learning algorithm, a result using the recommendation factor for each of the candidate events as input;
selecting and storing in the storage device an event among the plurality of candidate events based on the result of the calculating; and
presenting on a display the selected event.
8. The non-transitory computer-readable medium according to claim 7 , wherein the method further comprises inputting the recommendation factor into a feature vector for input to a recommendation algorithm that determines a possibility of the user executing the event.
9. The non-transitory computer-readable medium according to claim 8 , wherein the recommendation algorithm is one or more of a logistic regression algorithm, a support vector machine (SVM) algorithm, and an Extreme Gradient Boosting (XGBoost) algorithm.