Abstract:Author Profiling (AP) is a method of identifying the demographic profiles such as age, gender, location, native language and personality traits of an author by processing their written texts. The AP techniques are used in multiple applications such as literary research, marketing, forensics and security. The researchers identified various differences in the authors writing styles by analysing various datasets. The differences in writing styles are represented as stylistic features. The researchers extracted several style based features like structural, content, word, character, syntactic, readability and semantic features to recognize the profiles of the authors. Traditionally, the researchers extracted various feature combinations for differentiating the profiles of authors. Several existing works are used Machine Learning (ML) methods for predicting the author characteristics of a new author. The existing works achieved good accuracies for predicting the author characteristics by considering the both stylistic features and ML algorithms combination. Recently, in advent of Deep Learning (DL) techniques the researchers are proposed approaches to author profiling by using these techniques. Few researchers identified that the deep learning techniques performance is good for author profiles prediction than the results of style based features. In this work, a word embeddings based approach is proposed for gender and age prediction. In this approach, the experiment conducted with different word embedding models such as Word2Vec, GloVe, FastText and BERT for generating word vectors for words. The documents are converted as vectors by using the document representation technique which uses the word embeddings of words. The document vectors are transferred to three different ML algorithms such as Extreme Gradient Boosting (XGBoost), Random Forest (RF) and Logistic Regression (LR) for generating the trained model. This model is used for predicating the accuracy of age and gender prediction. The XGBoost classifier with word embeddings of BERT achieved good accuracies for age and gender prediction than other word embeddings and ML algorithms. The experiment implemented on PAN 2014 competition Reviews dataset for age and gender prediction. The proposed approach attained best accuracies for predicting age and gender than the performances of various existing approaches proposed for AP.

An Approach for Identifying Author Profiles of Blogs.

Weakly Supervised User Profile Extraction from Twitter.

An innovative framework for supporting content-based authorship identification and analysis in social media networks

Learning a Probabilistic Semantic Model from Heterogeneous Social Networks for Relationship Identification

A Multi-Source Integration Framework for User Occupation Inference in Social Media Systems

User Profiling Via Affinity-Aware Friendship Network

Social Network Extraction of Academic Researchers

FDHFUI: Fusing Deep Representation and Hand-Crafted Features for User Identification

A Graph-Based Author Name Disambiguation Method and Analysis via Information Theory

The collaborative search by tag-based user profile in social media

Key Technology Research on User Identity Resolution Across Multi-social Media

Studying User Footprints in Different Online Social Networks

Multiple userids identification with deep learning

A Unified Semi-supervised Framework for Author Disambiguation in Academic Social Network.

Social Media Profiler: Inferring Your Social Media Personality from Visual Attributes in Portrait.

User Identity Resolution Across Multi-Social Media

Social Network Integration: Towards Constructing the Social Graph

Identifying Topic Experts And Topic Communities In The Blogspace

A Word Embeddings based Approach for Author Profiling: Gender and Age Prediction

Author Name Disambiguation via Heterogeneous Network Embedding from Structural and Semantic Perspectives

Identifying User Profile by Incorporating Self-Attention Mechanism based on CSDN Data Set