Application for prediction of content’s popularity

Transcript

1. Application for prediction of content’s popularity Prepared by student of ICIT PI-417 Diachuk Marianna
2. Definition of the problem Users care how the content they produced is received by their audience. Predicting popularity allows users to analyse the ways of promoting their content and to come up with better and more efficient approaches of doing it. 2
3. Objective of work Effectively predict popularity of an Instagram post before posting it using Machine Learning anomaly detection algorithms. 3
4. Proposed solution The concept of predictive modeling can be applied with use of anomaly detection algorithms. Popularity is considered to be an anomaly. Model built with anomaly detection algorithm is trained to detect the anomaly in data. 4
5. Initial data 5 Data parsed from Instagram is in JSON format and requires further preprocessing.
6. Exploratory Data Analysis 6
7. Feature Engineering 7
8. Modeling The idea of modeling is to create a model of known relationships between features values in the dataset. The impact of each feature on the model score can be checked . 8
9. Model deployment as an application 9
10. Application Class Diagram 10
11. Model results Model performance metrics / Algorithms Anomaly detection with time series Decision Trees Logistic Regression AUC sample 1 0.8071264367 0.7057471264 0.7354022988 sample 2 0.8297458638 0.7819358690 0.7438001023 11
12. Thank you for your attention!