NLP Metrics Suitable for Various Categories of Image Captions

EasyChair Preprint 10803

Abstract

The aim of this project is to explore and identify suitable natural language processing (NLP) metrics for different styles and types of image captions. Image captions play an important role in providing additional context and information about images, and the effectiveness of the caption depends on various factors such as style, type, and category. The project will utilize a diverse set of image caption datasets from different domains, such as social media, news, scientific research, and other sources. Various NLP metrics, including sentiment analysis, readability, coherence, and topic modeling, will be applied to these datasets to evaluate their effectiveness in capturing the nuances of different caption styles and types. The project will analyze the effectiveness of each metric in capturing the nuances of different caption styles and types, such as humorous, informative, and poetic captions, as well as captions for different image categories, such as nature, sports, and food. The project will also investigate the impact of other factors, such as the length, structure, and content of the captions, on the applicability of NLP metrics. The findings of this project will contribute to the development of more accurate NLP models for analyzing different types of image captions, which will assist in improving the user experience of various image-based applications. Moreover, it will provide valuable insights into the relationship between the content of the image and the quality of the caption, enabling better captioning algorithms in the future.

Keyphrases: BLEU, CNN, METEOR, NLP, Rogue, SPICE

@booklet{EasyChair:10803,
  author    = {B Kezia Rani and Prajwal Reddy Dornala and Sravani Golla},
  title     = {NLP Metrics Suitable for Various Categories of Image Captions},
  howpublished = {EasyChair Preprint 10803},
  year      = {EasyChair, 2023}}