Detecting ideas in online communities : utilizing machine learning and text mining for finding ideas in online communities

Summary

Online communities serve as a gathering point for dedicated product users and consumers who discuss all imaginable topics. Scholars have argued that this discussion can lead to new ideas useful for firms. That is, if the ideas can be detected amongst the vast amount of information contained in online communities. The nature of online community data makes idea detection labour intensive and a systematic way of dealing with the data is needed if firms are to fully exploit online community ideas for innovation. This is the starting point for the research carried out in this doctoral project. The present doctoral thesis introduces an automatic method for idea detection aimed at screening large amounts of online community texts. The method is based on machine learning and text mining techniques and it is developed on two product cases related to brewing and Lego. The method relies on a large set of pre detected idea texts and non-idea texts for learning the lexical pattern embedded in idea texts. It is described how to pre-process the text data and how to adjust the machine learning techniques for optimal idea detection performance. Support Vector Machines and Partial Least Squares are used as machine learning techniques. The presented results show that when the method is trained for Lego idea detection and tested on an independent Lego hold-out set, the method obtains moderate to substantial agreement with human idea raters. When the method is trained for beer brewing idea detection on an independent hold-out set, the method obtains fair to substantial agreement with two brewing experts. Moreover the results indicate that people use specific idea words and expressions when they talk about ideas. This is why automatic idea detection is possible.

Read publication here:

NVA : hdl.handle.net/11250/2497664