Social media image classification using deep convolutional neural networks

Download

index.pdf

Date

2017

Author

Akpak, Çağrı Utku

Metadata

Show full item record

Item Usage Stats

164
views

128
downloads

Increasing popularity of social media platforms has led to an increase in the number of unclassified images. Given the complexity of images uploaded to these platforms and the number of classes available, it is clear that traditional image classification methods are not suitable for this kind of classification. Previous research on this topic primarily focuses on Deep Neural Networks to overcome the limitations of traditional methods. In these studies, researchers either limited the scope of their dataset; for example, handwritten digits, or combine their approach with Natural Language Processing methods to create meaningful descriptions. Similarly in this study, we use Deep Convolutional Neural Networks to classify social media images. Unlike previous approaches, there is no limitation on the scope of the images and classes represent textual tags that explain images in a simple and natural way. Moreover, the previous approaches do not allow class expansion after the training. To overcome this difficulty, a modular system is developed for classification. Separate networks are trained for each individual class and they are combined to create the overall system. Using this system new classes can be introduced without affecting the performance of the previously trained classes. Experiments are done on a dataset compiled from social media platforms and this approach achieves promising results.

Subject Keywords

Imaging systems., Image processing., Machine learning., Artificial intelligence., Neural networks (Computer science).

URI

http://etd.lib.metu.edu.tr/upload/12621351/index.pdf
https://hdl.handle.net/11511/26711

Collections

Graduate School of Natural and Applied Sciences, Thesis

Suggestions

OpenMETU
Core

Social media and word-of-mouth dispersion : factors affecting likelihood of diffusion Kursun, Zuhal Sinem; Yılmaz, Cengiz; Department of Business Administration (2013) This study aims to analyse the factors behind electronic word-of-mouth participation on social network sites, especially on Twitter. For this purpose, the general drivers behind such activity is classified, and other sub-factors were categorised under these drivers were categorised as influencers and determinants. Both individual and environmental factors were studied for a comprehensive analysis about factors increasing or decreasing the likelihood of diffusion of word-of-mouth messages. Besides the hypoth...
Collective classification of user emotions in twitter İleri, İbrahim; Karagöz, Pınar; Department of Computer Engineering (2015) The recent explosion of social networks has generated a big amount of data including user opinions about varied subjects. For classifying the sentiment of user postings, many text-based techniques have been proposed in the literature. As a continuation of sentiment analysis, there are also studies on the emotion analysis. Because of the fact that many different emotions are needed to be dealt with at this point, the problem becomes much more complicated. In this thesis, a different user-centric approach is ...
Determining user types from twitter account contentand structure Gürlek, Mesut; Toroslu, İsmail Hakkı; Department of Computer Engineering (2021-3-05) People are using social media platforms more and more every day; hence, they are be-coming suitable for research studies by their rich content. Twitter is one of the biggestand most widely used social media platforms, and many studies focus on Twitter forsocial media research. In this thesis, we propose methodologies for determining usertypes of Twitter accounts by their metadata, content, and structure. Our first problemis classifying organization vs. individual account types using only metadata. After weg...
Text classification in Turkish marketing domain and context-sensitive ad distribution Engin, Melih; Can, Tolga; Department of Computer Engineering (2009) Online advertising has a continuously increasing popularity. Target audience of this new advertising method is huge. Additionally, there is another rapidly growing and crowded group related to internet advertising that consists of web publishers. Contextual advertising systems make it easier for publishers to present online ads on their web sites, since these online marketing systems automatically divert ads to web sites with related contents. Web publishers join ad networks and gain revenue by enabling ads...
Combining topology-based & content-based analysis for followee recommendation on Twitter Yanar, Aysu; Karagöz, Pınar; Taşkaya Temizel, Tuğba; Department of Information Systems (2015) Twitter has become an important social platform for individuals and people share a high number of information about their personal lives, interests and viral news during emergencies. As of 2014, Twitter has 240 million active users and approximately 500 million tweets are shared every day. This information overload in Twitter has become a serious problem due to the growing volume of messages and increasing number of users. Recommender systems help to overcome this challenge. Finding interesting users and ge...

Citation Formats

Ç. U. Akpak, “Social media image classification using deep convolutional neural networks,” M.S. - Master of Science, Middle East Technical University, 2017.