General Online Research 2019
C05: GOR Thesis Award 2019 Competition: Bachelor/Master
Interactions on Twitter conducted at a cMOOC – Results of a mixed-methods study
Technische Universität Ilmenau, Germany
Relevance & Research Question. A massive open online course (MOOC) is an online course with open access and a theoretically unlimited number of participants. Since 2011, the potential of MOOCs and the associated problems have been discussed intensively. The public interest and the rapid dissemination of the courses underline their significance. Since the first MOOC in 2008, two different formats have been emerged: the behaviorist oriented xMOOC (extended MOOC) and the connectivist oriented cMOOC (connectivist MOOC).
This study deals with cMOOCs because - unlike xMOOCs - the format has a higher didactic potential. cMOOCs focus on shared, network based learning as well as on personal responsibility, motivation and commitment of the learners. Social media channels are integrated for sharing and networking. The interaction of the participants is considered to be a major factor for the success of a cMOOC. Nevertheless, there are few research results available in this area.
Against this background, the following research questions emerged:
Q1: What types of interaction occur on Twitter during a cMOOC?
Q2: What interactions occur at what stage of the learning process on Twitter during a cMOOC?
Q3: How are the interactions within the learning process on Twitter perceived by the participants and organizers of a cMOOC?
Methods & Data. Moore’s three types of interaction and Salmon’s five-stage model of teaching and learning online were applied as a theoretical framework. Moore differentiates between three types of interaction: “learner-learner”, “learner-instructor”, and “learner-content” interaction. The first two types describe the interactions between learners or between learners and teachers. The interaction between learner and content refers to the individual’s interaction with the content.
The second model of the framework is the five-stage model of teaching and learning online according to Salmon. The basic idea is that the learners go through an individual learning process, which is divided into five stages: The first level is the stage of access and motivation and is about setting up processes and motivating learners to participate. The second stage is the stage of online socialization and is characterized by the formation of a community. In the third stage, the stage of information exchange, a common understanding among the participants is built up. This is the basis for the fourth stage, the stage of knowledge construction. In the fifth stage of development, learners reflect the acquired knowledge.
In a mixed-methods study it was analyzed which type of interaction occurred during the process of learning and how it was perceived by learners and organizers of a cMOOC. For that purpose n=2,497 Twitter messages of a partial sample were explored by a quantitative content analysis in 2018.
Based on the findings gained from the content analysis, an interview guideline was developed. Eight telephone interviews with two organizers and six learners were conducted. The sample of the learners was chosen considering the characteristics gender and activity level. The activity level was derived from the content analysis. Three male and three female persons were interviewed, the male and the female group each consisted of one person with a low, one with a medium and one with a high level of activity.
Results. The analyzed Twitter messages were published by 170 different users. 88% of the Twitter users were identified as learners, 8% as weekly hosts and 4% as organizers. 76% of the messages were published by the learners, 21% by the organizers and 4% by the weekly hosts (n=2,497).
The findings revealed that the most interaction occurred between learners on Twitter during the cMOOC. In the first stage of the learning process, the interaction between learners and organizers was the highest. The relevance of “learner-instructor” interaction at the beginning of the course was affirmed by this and other studies. In the second, third and fourth stage, interaction between the learners was most common. In the fifth stage, mainly the “learner-content” interaction took place. Twitter was used primarily for the second and third stage of the learning process – for online socialization and information exchange. The short message service was hardly used for the discussion and reflection of the content, which corresponds to the fourth and fifth stage of the learning process.
Overall, the participants interacted relatively often in their Twitter messages. Unlike in other MOOCs, the interactions – especially the interaction between the learners – did not decrease continuously over time. The study shows that the organizers of the cMOOC interacted regularly and very often with the participants. Furthermore, the presence of weekly hosts had no effect on the quantity of learners’ interactions.
Added Value. As the explorative study focuses on a particular case, the results can’t be generalized into being applicable to all cMOOCs, but the study provides a unique insight into an isolated instance of interactions within the learning processes on Twitter. The finding that can be of general value is the knowledge on cMOOC learner’s behavior and interactions on Twitter. MOOC designers and facilitators can consider these results in the conception and moderation of future MOOCs and thus can better engage with their learners.
Can these stars lie? Online reviews as a basis for measuring customer satisfaction
Hochschule für Technik Stuttgart, Germany
Relevance & Research Questions:
Consumers who shop online find great value in the opinions and recommendations of other consumers when searching for product-related information. Online reviews written by customers in online shops or review sites therefore influence readers and consequently the success of products or services. Online reviews can be interpreted as the behavioral result of the experienced level of customer satisfaction. Assessing customer satisfaction is of major interest for companies, which is why product managers and other executives rely on various sources and measurements to gain insights from their customers. One method - that has been developed in recent years to gain more insight into product-related customer satisfaction - is the mining and monitoring of online reviews using automated software solutions.
However, studies have shown that these reviews might be compromised by biases affecting the customers writing them. Customers who purchased a product tend to have an overall positive attitude towards this product because they consciously decided to buy it, resulting in a more positive rating (purchasing-bias; Hu, Pavlou & Zhang (2007)). Furthermore, the distributions of star ratings, the most common way of rating satisfaction with a product on platforms like Amazon, are not normally distributed. Instead they often follow a “J-shaped” distribution which means that online reviewers rate products rather extremely than moderately. Research suggests that this is the result of a self-selection of customers moderately satisfied with the product. Moderately satisfied consumers lack motivation, like excitement or disappointment, to write an online review (under-reporting bias; Hu et al. (2007)).
The described shortcomings of online reviews lead to the question whether they are an appropriate source for companies to derive customer satisfaction from. To approach this issue in the current study, the method of review monitoring was compared to the method of customer satisfaction surveys, a widely established method of measuring customer satisfaction. The study followed two research questions:
1. Can online reviews give a realistic image of the product-related customer satisfaction?
2. What is the added value of review monitoring compared to online surveys and how can the two methods complement each other?
Methods & Data:
Washing machines were chosen as research subject to compare the two methods of satisfaction measurement, an object that is available in any household and is usually bought after an extensive search for information. The data basis was limited to six washing machine brands in order to increase the comparability of the data.
The study followed a four-step research process. First, a code system was developed in order to categorize all relevant evaluation criteria for washing machines. The data basis were 300 online reviews for washing machines. Eight superordinate performance parameters could be identified in the review texts, e.g. washing, price and customer service. The overall satisfaction assessment corresponds to the five-star scale used in all reviews.
Second, a questionnaire based on the code system developed in the first step was designed in order to measure the product-related customer satisfaction. The questionnaire contained open-ended questions as well as five-level scales to assess the satisfaction with the washing machine in general and its performance parameters. Additionally, the relevance of assessment criteria was measured. The data was collected via an online survey. The sample consisted of 510 washing machine users who purchased a washing machine between 2016 and 2018 and it was composed of 45% women and 55% men with an average age of 45 (SD = 12.4).
Third, a sample of 589 consumer reviews was drawn and coded with the previously developed code system. The sample was composed of 57% women, 36% men and of 7% authors with unknown gender. The open-ended questions from the online survey were coded with the help of the same code system.
In the last step, the data retrieved from the user reviews and from the satisfaction survey were analyzed and compared on a quantitative and qualitative level.
Results of the analysis of the overall satisfaction suggest that authors of online reviews are slightly more satisfied with their washing machines than the survey participants. But the differences in ratings are only very small. Both the distributions of ratings from online reviews and the survey are “J-shaped”, meaning they are not normally distributed but rather polarized. However, online reviewers rate their washing machines more often with extreme values on the satisfaction scale than the participants of the survey. These findings support the existing research on the under-reporting bias.
Analyses of assessment criteria and performance parameters show that authors of online reviews mention more distinctive criteria in their review texts than participants of the surveys do in open-ended questions. In addition, authors of online reviews mention criteria that are perceived as important more often than the participants of the survey. This indicates the additional qualitative value of online reviews.
The qualitative analysis shows that online reviews can provide a genuine insight into the usage of the appliances and into the living environments of customers. The distinct performance parameters are rated similarly by authors of online reviews and participants of the survey, which suggests a consistency in more detailed assessments.
The study shows the advantages and disadvantages of both methods to measure customer satisfaction. Results show that online reviews are indeed biased to some extent. The distribution lacks moderate ratings that are more common in customer satisfaction surveys. This can lead companies to over- or underestimate the customer satisfaction with their products when only referring to online reviews. On average, products are rated very similarly with both methods, which speaks for their reliability. In terms of qualitative insights into the use of products or product features that can disappoint or excite customers, online reviews can be a valuable source for companies e.g. in product management or development.
Overall, the method of review monitoring appears to be a valuable tool for companies to gain a real-time overview of customer feedback and satisfaction with their product portfolio. Especially for companies that offer various product categories with different models, it is a way to track problems systematically and quickly in a cost-effective and continuous manner. It is recommended to use additional market research methods, such as customer satisfaction surveys, to balance biases and to validate the qualitative input gained from online reviews with more quantitative measures. The mix of methods can provide a comprehensive view of product-related customer satisfaction.
Comparing the Portrayal of German Politicians in Bing News and Google News Search Results
Technische Universität Ilmenau, Germany
Relevance & Research Question:
The internet is crucial for keeping up with news: Almost 60 percent of the German population above 14 years access online news (van Eimeren & Koch, 2016). The majority of German internet users rely on (news) search engines to navigate online information (van Eimeren & Koch, 2016). By filtering the available (news) content and deciding which content to present to the users, these news aggregators act as secondary gatekeepers (Nielsen, 2014; Singer, 2014; Wolling, 2005). Editorial lines and selection criteria of (human) primary gatekeepers are generally known, but there is a lack of information about the workings of news search engines’ selection and ranking algorithms (Lewandowski, 2015).
The findings of several US-American studies indicate that there may be relevant differences between different news search engines. These observed differences point to two dimensions: the variety of presented news sources , and the portrayal of politicians on a content level. Compared to Yahoo News, the search results of Google News showed a larger variety of sources and less concentration on big publishing companies (Bui, 2010). Search results for the terms “George W. Bush” and “John Kerry” retrieved by Yahoo News were more neutral, while Google News presented more explicitly judgmental news articles (Ulken, 2005). Thus, by relying on these services users might – without being aware of this – receive differing information about the same subjects (Bozdag & van den Hoven, 2015).
However, there are considerable research gaps - especially in the context of specialized news search engines. First, there is a lack of comparisons between different news search engines in the literature. Most studies focus on only one service or compare search engines and other services (e.g. news portals). Second, only very few studies compare the search results on a content level. Third, the few comparative studies that consider the retrieved content are all rather old and focus on the US. These past findings may no longer apply and are not necessarily valid for the German editions of the news search engines.
This study aims to close these research gaps. In addition to a descriptive examination of the retrieved sources, the study focuses on the portrayal of politicians in two popular news search engines in Germany. Two dimensions of portrayal are considered for this comparison: First, the portrayal of politicians’ private lives (privatization), which includes information on lifestyle, families, and friends (Holtz-Bacha, Langer & Merkle, 2014). Second, the portrayal of politicians’ professional characteristics as leaders (leadership images), such as political skills, vigorousness, and charisma (Aaldering & Vliegenthart, 2016).
The study is guided by the following research questions:
To what extend does the portrayal of German politicians differ between Bing News and Google News results?
SRQ1 Are the two news search engines’ results favoring certain politicians by portraying them in a better light than others?
SRQ2 Are the two new search engines’ results favoring certain political parties by portraying their politicians in a better light than others?
Additionally, this study tries to identify potential differences in the types of portrayals themselves:
SRQ3 Are the two news search engines focusing on professional or private information about the politicians?
SRQ4 Are the two news search engines emphasizing different dimensions of leadership in their portrayal of politicians?
Method & Data:
The empirical assessment of search engine algorithms is challenging, as researchers can only compare outputs for identical inputs to infer potential differences between services (Lewandowski, 2015). Thus, this empirical examination followed a three-step process. First, 20 search queries with 16 different search terms (names of politicians and political parties, current political issues) were sent simultaneously to the selected news search engines between December 2017 and January 2018. The first search result page was archived for each query and the first ten entries were considered for analysis. Second, the first five news articles per result page were accessed and archived. To avoid unwanted personalization based on the accessed news sources, the first step was fully completed before starting the second. Third, the resulting dataset of 400 search results and 200 full-length news articles was examined via quantitative content analysis. The analysis focused on the variety of presented sources and the portrayal of the three first-mentioned politicians per article. The portrayals were coded as positive, negative, or neutral. In total, 20 categories on the portrayals of politicians – 8 for privatization (Holtz-Bacha et al., 2014) and 12 for leadership images (Aaldering & Vliegenthart, 2016) - were considered.
SRQ1: There are no signs of a systematic bias for or against specific politicians. However, there is a lot of variance in the portrayals and there are two individual cases in which the portrayals of the same politicians differ significantly between the search engines. Additionally, the portrayals in Google News results are more likely to be negative (42%) or positive (28%) than portrayals in Bing News results (negative: 37%, positive: 17%, χ² = 9.821, p = .007).
SRQ2: There are no indicators for systematic bias for specific political parties with regard to the portrayal of associated politicians.
SRQ3: Regardless of the selected search engine, information about politicians’ private lives play almost no role in the observed portrayals of politicians.
SRQ4: There are no clear indicators of the search engines emphasizing different dimensions of leadership.
Additionally, there are significant differences in the selected news sources: Bing News results contained more online-only media sources (25%, Google: 8%, χ² = 23.91; p < .001) and wire reports (52%, Google: 27%, χ² = 13.10; p < .001). Google News retrieves more broadcast sources (16%, Bing: 9%, χ² = 23.91; p < .001) and – at least in this sample – more opinion articles (12%, Bing: 6%, χ² = 2.20; p =.138).
Although no systematic bias considering the portrayal of politicians is evident, the choice of (news) search engine influences the content that will be presented to the users. The findings indicate that algorithmic bias might manifest in different dimensions than expected. Instead of a political left-right bias, the significant differences in the presented sources and in the portrayals of some politicians show that new forms of bias should be considered in future research.
Citizens also need to be educated about secondary gatekeepers as part of code literacy: The commonly cited advice to cross-reference information between different sources might have to be expanded to also include cross-referencing between different services to find these sources of information.