Yapay Zeka Hızlı Bakış & Çıkarımlar
AI InsightIn the next stages of the project, innovative methods used in this field such as “Collaborative Filtering” and “RNN” will be performed together with the “ClickStream” data of the customers and the success rate will be tried to be increased.
📌 3 Maddede Bu Makale
Abstract
Armut Technology is an online platform that brings together customers and service providers, and positions service providers as business partners with the principle of "Crowdsourcing". Nearly 4000 services are offered within the company. This number is increasing gradually as new service requests are also received from service providers. When customers login to the website or application, they search from the wide service pool by typing their desired service description. With this project, it is aimed to provide the services they need in real time when they are online by predicting them with artificial intelligence supported algorithms. The related topic is modeled as “Recommendation Engine” under the machine learning discipline. All service requests coming in 2020 were used as a training set. Since the queues of the services requested by the customers are interconnected in terms of temporality, the requested service queues are modeled according to the "Conditional Probability Based Prediction" method. In order to capture exceptional customer behaviors, customer specific habits have also been added to the service list. All machine learning models run on the AWS cloud ecosystem. It has been developed with the principle of running web services in Docker containers, which is the industry standard and used during the service of machine learning models to the customer. “Top-8 Service Accuracy” was chosen as the success metric of the project. The success rate of 22%, which is currently achieved by combining popular services throughout Turkey, has been increased to 37% with the new algorithm supported by AI. This rate comes up to 44% when we look at the customers who have had at least 1 service request in the past. The current service recommendation system, which has difficulty in capturing special customer behaviors with the perspective of “popular service throughout Turkey”, has significantly improved with the new AI supported approach by taking into account customer habits and the relationship between services. In the next stages of the project, innovative methods used in this field such as “Collaborative Filtering” and “RNN” will be performed together with the “ClickStream” data of the customers and the success rate will be tried to be increased.