MACHINE LEARNING METHODS FOR FORECASTING OF SOCIAL NETWORK USERS’ REACTION

Subject of Research. The paper presents comparative analysis of forecasting methods for users’ response to a message emerged on social media and application of machine learning. The method that provides the highest accuracy is preferred. Method. The following machine learning methods were applied: s...

Full description

Saved in:
Bibliographic Details
Published in:Nauchno-tekhnicheskiĭ vestnik informat͡s︡ionnykh tekhnologiĭ, mekhaniki i optiki mekhaniki i optiki, 2020-01, Vol.20 (1), p.118
Main Authors: Popova, E P, Leonenko, V N
Format: Article
Language:Russian
Subjects:
Acceleration
Containment
Digital media
Documents
Forecasting
Machine learning
Multilayer perceptrons
News
Regression analysis
Social networks
Statistical analysis
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Subject of Research. The paper presents comparative analysis of forecasting methods for users’ response to a message emerged on social media and application of machine learning. The method that provides the highest accuracy is preferred. Method. The following machine learning methods were applied: support vector method, gradient boosting methodrandom forest and multilayer perceptron. The statistical method of regression analysis — linear regression — was used as a reference forecasting method. Vectorization of records for obtaining quantitative characteristics of their content was carried out using such methods as: “Bag of Words”, TF-IDF and Word2Vec. The forecast quality was evaluated by R2 determination coefficient. Main Results. A numerical experiment was performed using a data set collected on the VKontakte social network. The set contained information about community subscribers, publications, “I like” and “tell friends” marks and comments on publications. The number of marks and comments under the posted publication was projected, depending on its content. The most accurate results were obtained when predicting the number of comments. The quality of forecasts for the number of “I like” and “tell friends” marks turned out to be lower. Practical Relevance. The results of the work can be used in analyzing the effect of various news, including fake news, on users of social networks. The development of forecasting methods provides the planning of measures for acceleration or containment of the messaging distribution.
ISSN:2226-1494
2500-0373
DOI:10.17586/2226-1494-2020-20-1-118-124