The paper 'Connected scholars: Examining the role of social media in research practices of faculty using the UTAUT model' [3] uses qualitative methods in form of interviews to gain a deeper inside into the way of how social media is used for research. To ensure to get a representative group of participants, they selected people with different nationalities and working positions. I believe that they were able to achieve that goal even though they recruited only people who are part of a specific society.
At the beginning of the interview they were shown a list of different social medias and had to answer questions about it. It is highly probable that participants were biased by these list.
Drawbacks of interviews are often that the interviewer has to react spontaneously according to the participants answers. But since the researchers in this study seem to have prepared themselves and followed a kind of interview guideline, this disadvantage is kept under control.
I found it interesting to read about an anaylsis approach they used to evaluate the qualitative data. I did not knew that detailed evaluation models with coding schemata are available for qualitative research - especially interviews.
It was an appropriate choice to use an interview as a qualitative method in this study. It was suitable to answer the research question and I would not consider quantitative methods as a necessary completion in this case.
To put it in simple words, a case study is a research theory which results into qualitative and/or quantitative data. A certain problem or a set of problems that should be resolved and solution approaches are evaluated. Beforehand, a detailed description of the research question should be defined. This is necessary, because case studies often result in a huge amount of data which could cause the research focus to become blurred [3].
I choose the paper 'The contribution of context information: A case study of object recognition in an intelligent car' [2] from the Neurocomputing journal. They tried to investigate possibilities of learning on a higher abstraction level. This machine learning strategy is applied to autonomous agents which are used in cars to recognize other vehicles. Their case study consisted of the comparison of different learning algorithms under real-world influences. By evaluating the results of the study they concluded that learning algorithms lead to a better performance in object recognition.
In my opinion they defined the research question very well. Since there are a lot of papers from related research fields which reported good results using learning on a higher abstraction level, it makes sense to test this strategy for object recognition. The specific problem definition was the hypothesis that the use of these algorithms would minimize the false negative rate at the costs of a low true positive rate of object recognition. This means that they expected the algorithms to reduce the number of missed object whereas they did not take the falsely detected objects into account. In my point of view, this is a legitimate and verifiable problem definition.
Moreover, they clearly defined their requirements for the learning algorithms. It appears to me, that they carefully chose the suitable algorithms and that they did a lot of research to be able to get a sufficient selection.
To investigate the performance and the effort of the algorithms they used SamSys to simulate different road environments and HRI RoadTraffic as a database. This part of the paper could be a bit more detailed, it lacks an explanation why they chose to use these tools and whether there might have been some alternatives.
The cases of this study are different streams from a road dataset. The streams have only been selected if there were vehicle annotations available. They did not seem to had other criteria to choose a certain stream as a study case. I guess that the study would have benefited from more specific selection criteria. But at least they tried to get road streams with different weather conditions.
In order to measure the data, they created a baseline of object detection results from a visual classification system. The data they collected using the different learning algorithms were compared to that baseline. They tried to reach closure by comparing the algorithms only to this baseline and by building clusters of cases. That means that they divided the dataset into cases with a rainy weather condition and the ones with standard weather condition. I think that this was a sufficient way of grouping the data and come to meaningful results.
In general I would they, that it was not easy to follow the study even though I have some background knowledge in machine learning and autonomous agents. But as fas as I can see, the study is well designed and they analysed the data according to their initial problem definition.
References:
[1] Eisenhardt, K. M. (1989). Building Theories from Case Study Research. Academy of Management Review, 14(4), 532-550.
[2] Gepperth, A., Dittes, B., & Ortiz, M. G. (2012). The contribution of context information: a case study of object recognition in an intelligent car. Neurocomputing, 94, 77-86.
[3] Gruzd, A., Staves, K., & Wilk, A. (2012). Connected scholars: Examining the role of social media in research practices of faculty using the UTAUT model. Computers in Human Behavior, 28(6), 2340-2350.
No comments:
Post a Comment