Crowdfunding Campaigns. Success Prediction Through Natural Language Processing

Master's Thesis , 2021 , 83 Pages , Grade: 1,0

Business economics - Company formation, Business Plans

This thesis examines whether a data mining approach, such as natural language processing, can help the founders of crowdfunding campaigns be more successful.

In a data mining framework 493,324 campaigns of the two popular crowdfunding platforms Kickstarter and Indiegogo were analyzed by natural language processing using different artificial neural networks to obtain the information needed by the founders. For frequently occurring categories, a reliable classification of the category was possible. For rare ones it was less precise. It was also shown that the more a founder concentrates on a specific category when setting up a campaign, the more likely it was that a campaign would be successful. A prediction of campaign success was also possible but was influenced by the nature of the data set. It was demonstrated that this approach could generate important information that could lead to a competitive advantage of the founders for most of the campaigns in the dataset.

Crowdfunding is an emerging industry which has gained considerable attention in recent years. Competition among campaigns and founders will therefore become increasingly intense. This means, that founders must gain a competitive advantage over the competitors to be successful. Data mining approaches which also include natural language processing could be suitable to assist the founders with valuable information when setting up campaigns and thus enable them to gain a competitive advantage. Especially the right categorization on a crowdfunding platform and prediction of success are important information to support the founders.

Inhaltsverzeichnis (Table of Contents)

• Abstract
• Introduction
• Theoretical Foundation
• Data Mining Process
• Natural Language Processing
• Definition and Distinction
• Preprocessing and Feature Selection
• Vectorization of Words
• Architecture and Types of Artificial Neural Networks
• Feed Forward Neural Networks
• Convolutional Neural Networks
• Graph Neural Networks
• Recurrent Neural Networks
• Methods for Performance Evaluation
• Applied technologies
• Methodology
• Business Understanding
• Data Understanding
• Data Preparation
• Modeling
• Convolutional Neural Networks
• Graph Neural Networks
• Bidirectional LSTM
• Evaluation of Results
• Convolutional Neural Network Results
• Category Prediction by Convolutional Neural Network
• Success Prediction by Convolutional Neural Network
• Graph Convolutional Neural Network Results
• Category Prediction by Graph Convolutional Neural Network
• Success Prediction by Convolutional Neural Network
• Bidirectional LSTM Results
• Category Prediction by Bidirectional LSTM
• Success Prediction by Bidirectional LSTM
• Discussion

Zielsetzung und Themenschwerpunkte (Objectives and Key Themes)

This thesis aims to examine whether information regarding the categorization and success prediction of crowdfunding campaigns can be generated using natural language processing (NLP) and data mining approaches. The goal is to provide valuable insights for founders setting up campaigns on platforms like Kickstarter and Indiegogo, enabling them to gain a competitive advantage.

• The application of NLP and data mining for analyzing crowdfunding campaigns.
• The use of artificial neural networks for campaign classification and success prediction.
• The influence of campaign category and founder focus on campaign success.
• The potential of data-driven approaches for supporting crowdfunding success.
• The challenges and limitations of applying NLP and data mining to crowdfunding data.

Zusammenfassung der Kapitel (Chapter Summaries)

The thesis delves into the theoretical foundation of data mining and NLP, exploring key concepts such as the CRISP-DM model, preprocessing techniques, vectorization methods, and various types of artificial neural networks. The methodology chapter details the data collection and preparation process, outlining the specific steps involved in utilizing convolutional neural networks, graph neural networks, and bidirectional LSTMs to achieve the desired categorization and success prediction objectives. The evaluation of results section provides a comprehensive analysis of the findings for each model, discussing the accuracy, performance, and limitations of each approach.

Schlüsselwörter (Keywords)

Crowdfunding, Natural Language Processing, Artificial Neural Networks, Data Mining, Campaign Classification, Success Prediction, Kickstarter, Indiegogo, Competitive Advantage, Data Analysis, Machine Learning, Deep Learning, Text Classification, Sentiment Analysis, Data Set, Campaign Focus, Category Influence.

• College: Hamburg University of Technology (Institute of Entrepreneurship)
• Grade: 1,0
• Author: Benjamin Brummer (Author)
• Publication Year: 2021
• Pages: 83
• Catalog Number: V1382124
• ISBN (PDF): 9783346931719
• ISBN (Book): 9783346931726
• Language: English
• Tags: Deep learning, Neural network, Neural networks, Crowdfunding, Kickstarter, Indiegogo, Graph Convolutional Neural Network , GCN, Long Short-Term Memory Neural Network , LSTM, Convolutional Neural Network , CNN, NLP, Artificial Intelligence, AI, Phyton, R, Keras, Tensorflow