Concept Drift Detection From Text Using Graph Resolution

Abstract : Machine learning models that track how data changes over time can have problems when the data patterns themselves change. This change in data can make the models work less well. It's hard for models to handle these changes, especially when dealing with complex text. Traditional methods for detecting these changes use simple statistics and check model performance, but they often miss important details in text data. This paper suggests a new way to find these changes by turning text into graphs. Graphs help us see how words and ideas are connected and how these connections change over time. We use a special graph model that watches how the graph grows and how the connections within it change. We look at things like which words are most important, how groups of words are related, and how strong the connections between words are. This helps us see when the data is changing. We tested this method on different kinds of text data and compared it to older methods. The results show that using graphs to look at text is a good way to find changes in the data. This method is more accurate and makes the models more stable.

Keyword Concept Drift Detection , Text Mining , Graph Resolution ,Machine Learning, Natural Language Processing (NLP) , Data Streams