Anomaly detection in online social media

Title

Anomaly detection in online social media

Creator

A K, Sujatha.

Contributor

K,Balachandran .

Description

Online Social Media (OSM) is a platform where users post opinions, discussions, product reviews, random thoughts, advertisements, comment exchanges and status updates.These platforms help in text mining applications such as prediction of election results, newlinestudying global mood trends, public perception of a national concern or an issue, mining of public health knowledge, detecting epidemics and business analytics. These newlineapplications also present some research challenges like personal data stealing, community phishing, hate speeches, spreading misconceptions, cyber bullying and terror attack planning. Some of these challenges are anomalies or outliers which don t conform with the majority ones. The anomalies focused in this research work are behavioral and content anomalies. Data preprocessing for textual data from OSM plays an important role for creation of the Vector Space Model (VSM) which is used as an input for behavioral and content anomaly models. The contents posted by the public in OSM is written using natural language and sometimes may not follow the formal communication mode. It has lexical, newlinesemantic and syntactic ambiguities and becomes a challenging task to extract accurate information and discover logical patterns during the text mining process. Some of the commonly used methods for text mining are, Bag of Words (BoW), N-grams and Term newlineFrequency-Inverse Document Frequency (TF-IDF). Few limitations of these techniques newlineare, high dimensional sparse feature vectors, missing contextual meaning, presence of newlineweak features and Part of Speech ambiguity. In this research study, an improvised Feature Engineering model is proposed which is a combination of Forward Scan Trigrams and weighted TF-IDF to address the creation of an efficient Vector Space Model (VSM). This proposed model is used with an improvised Feature Hashing technique to address the removal of weak features.

Source

Author's Submission

Date

2021-01-01

Publisher

Christ(Deemed to be University)

Subject

Computer Science and Engineering

Rights

Open Access

Relation

No Thesis

Format

PDF

Language

English

Type

PhD

Identifier

http://hdl.handle.net/10603/426659