J. Žižka, O. Polcer: Comparison of methods evaluating results of clustering Diploma thesis, Mendel University in Brno, 2015. This thesis describes in detail data clustering, development of own clustering application, its comparison with the programme Cluto and analysis of results.

Firms in highly dynamic environments focusing on innovation in their products and services, often encounter elevated amounts of uncertainty regarding the future direction of technological change. Finding reliable and imbedded information enhances a firm’s ability to tackle new markets and take advantage of possible hidden opportunities. To reduce uncertainty, obtain hidden knowledge, and gain competitive advantage, environmental scanning, which is one of the main components of foresight, is recommended by scholars of strategic management. The process of detecting weak signals for shedding light what one authority calls “blurry future zones” (Day & Schoemaker, 2005, p.1) has currently been receiving attention in environmental scanning studies. Some studies emphasize the importance of the subject; yet they offer few practical methodologies for actual cases. To help address this gap, this research introduces a new approach for detecting weak signals during Internet-based environmental scanning by applying the Cluto toolkit (see Section 4.7) plus using human judgment. This novel methodology is applied to the application of Micro Tiles, a recent innovative product of a digital display company located in Ontario, Canada, Christie Digital Company. In the conduct of this exploratory research, about 40,000 HTML pages were retrieved from the Internet in a search during 2009. To extract weak signals information from the retrieved unstructured texts, documents were grouped into a number of clusters by the CLUTO software. Two subject matter experts compared and evaluated the cluster results for the purpose of finding potentially relevant information in regard to the company’s strategic intent. Analyzing the clusters, the experts reduced the number of clustered documents from the original corpus into smaller sets with the goal of finding more relevant and unexpected documents (weak signals). The relevancy and expectedness of information in documents were two measurements as related to weak signals. The trends of the study indicate that as anticipated both experts found more unexpected documents in the smaller sets rather than the larger ones. Moreover, regarding one expert’s analysis, the smaller sets contain documents that are more relevant to the domain of interest. Overall, according to one expert, documents existing in the smaller sets display more weak signals. This emerging methodology offers a practical procedure to apply web-based information in the development of a company’s environmental scanning procedures. Using this methodology, managers can employ both computer tools and human sense-making methods to detect potential weak signals and reduce certain biases in the detection process.

Firms in highly dynamic environments focusing on innovation in their products and services, often encounter elevated amounts of uncertainty regarding the future direction of technological change. Finding reliable and imbedded information enhances a firm’s ability to tackle new markets and take advantage of possible hidden opportunities. To reduce uncertainty, obtain hidden knowledge, and gain competitive advantage, environmental scanning, which is one of the main components of foresight, is recommended by scholars of strategic management. The process of detecting weak signals for shedding light what one authority calls “blurry future zones” (Day & Schoemaker, 2005, p.1) has currently been receiving attention in environmental scanning studies. Some studies emphasize the importance of the subject; yet they offer few practical methodologies for actual cases. To help address this gap, this research introduces a new approach for detecting weak signals during Internet-based environmental scanning by applying the Cluto toolkit (see Section 4.7) plus using human judgment. This novel methodology is applied to the application of Micro Tiles, a recent innovative product of a digital display company located in Ontario, Canada, Christie Digital Company. In the conduct of this exploratory research, about 40,000 HTML pages were retrieved from the Internet in a search during 2009. To extract weak signals information from the retrieved unstructured texts, documents were grouped into a number of clusters by the CLUTO software. Two subject matter experts compared and evaluated the cluster results for the purpose of finding potentially relevant information in regard to the company’s strategic intent. Analyzing the clusters, the experts reduced the number of clustered documents from the original corpus into smaller sets with the goal of finding more relevant and unexpected documents (weak signals). The relevancy and expectedness of information in documents were two measurements as related to weak signals. The trends of the study indicate that as anticipated both experts found more unexpected documents in the smaller sets rather than the larger ones. Moreover, regarding one expert’s analysis, the smaller sets contain documents that are more relevant to the domain of interest. Overall, according to one expert, documents existing in the smaller sets display more weak signals. This emerging methodology offers a practical procedure to apply web-based information in the development of a company’s environmental scanning procedures. Using this methodology, managers can employ both computer tools and human sense-making methods to detect potential weak signals and reduce certain biases in the detection process.

The first part of thesis focuses on theoretical introduction to the methods of text mining -- Information retrieval, classification and clustering. LSA method is presented as an advanced model for representing textual data. Furthermore, the work describes source data and methods for their preprocessing and preparation used to enhance the effectiveness of text mining methods. For each chosen text mining method there are defined evaluation metrics and used already existing, or newly implemented, programs are presented. The results of experiments comparing the effects of different preprocessing type and use of different models of the source data are then demonstrated and discussed in the conclusion.

The diploma thesis focuses on unstructured textual data preprocessing in relation to text mining. A series of experiments oriented to text mining is designed and carried out. The effect of different techniques of textual data preprocessing to the entire text mining process and its results is evaluated based on output of the experiments.