• Tidak ada hasil yang ditemukan

Table 4: Results of T-Test paired samples where mean, standard deviations, t-test values, and its p-values on VTComp and Sketch.

Variable VTComp Sketch t p-value

M SD M SD

Easy to use 5.4 0.96609 4.1 1.19722 -2.751 0.022 Easy to learn 5.7 0.67495 2.8 1.03280 -6.692 0.000 Perceived pleasure 5.7 0.82327 3.2 1.22927 -6.708 0.000 Perceived endorsement 5.4 0.96609 2.6 0.96609 -7.203 0.000

p = 0.000. Overall, there was a significant difference between VTComp and sketch group on the post survey questions response by participants as shown in Table 4.

Further, we observed that participants refer to textual information view on several occasions right after the target document was processed. Subsequently, participants selected the elements from the graphical response for the visualization process. This was comparable with the quali- tative analysis results in which we found that the most comfortable feature was the extraction of data as they interacted with textual information and a graphical view at the starting of the visual creation process. These results are presented in Figure 21.

Figure 21: How participants interacted with different views, such as Process bar and chart component control board, Annotation tools view, Artboard view, Textual information view, Graphical response view, and Annotation control board view while task was performing.

Important Textual Information is preferred over full article read From the data, we inves- tigated that participants preferred to read the summary after processing the data, important keywords, and went through headings instead of reading the article. Although the article reading proceeding point was available in VTComp. For instance, Participant P5, P11, P12, P14, P15,

P16, and P7 used the textual information view to get an overview of the article as shown in Figure21. Conversely, some participants(P7, P8, and P10) preferred to read the articles.

UX could be worse when it caused irritation

During the experiment, we perceived that the annotation control board merely interrupted the participants when they wanted only one-time use of some annotation elements, such as circle shape. Currently, VTComp opens the annotation control board every time whenever a user selects the shape. As participant P10 notes:” It is annoying that it opens when we select the element every time, I think it should be closed until we want to open it again." P10 further suggested that there should be an event from where we open this window manually.” Even though the rest of the participants did not object to it, but it can be the noticeable point that may interrupt the user while designing, so we considered this and give the option which will be controlled by the user.

Although the accuracy of extraction is not that good, still the modifying the data can be enough for scaffolding and for curating informative visualization. The survey results also depict this as the feedback score was quite excited and positive (easy to learn - 5.55, easy to use - 5.8, enjoying generating - 5.85 visualization and will recommend to others - 5.7, these results were obtained using 7 points Likert measurement scale ranging 1 to 7 in agree to disagree order respectively)

Figure 22: Participants overall click interaction with VTComp elements during experiment.

The Figure22 shown presented participants’ interaction with various views of VTComp over the time when they were designing the visuals, which indicate that while designing the infor- mative thumbnails or visuals the usage of the Annotation view was high, which was followed by article top bar and chart component board view.

Qualitative Analysis Results:

The qualitative data from the think-aloud sessions and interviews were transcribed and analyzed using Strauss and Corbin’s version of ground theory [112]. The analysis consisted of three stages.

1) Open coding 2) Axial coding and 3) Selective coding. Preliminary, we analyzed and discuss the interview response data and shared ideas. We observe the main concepts and contradictions

in the first stage. In the second stage, we performed the linking between concepts to organize them into relevant categories by using keywords tagging. Finally, in stage third, we analyzed all the categories generated from concept data and integrated them into three core categories.

With the purpose of investigation of users conceived in more closely and finding the pattern behind the semi-structured interview respond, we used the qualitative analysis technique(which discussed in the above paragraph). Particularly, we sought to identify participants’ perceptions to interact with the system and its features importance. We identified that participants want some of the elements should automatically open as guidelines when they select. Next, they also want to enable the processing of PDF files using this tool. In addition, some of the participants try this tool with their own research area’s live documents and they were excited and comment that VTComp can be very useful for quick visualization creation, which they can further used for their meetings. Overall, Participants concede that it was easy to use and time-saving tool for creating informative visualization from live documents.

Figure 23: Bar chart displays the occurrence frequency of VTComp’s comfortable feature in interview session.

Figure 24: Bar chart presents the qualitative results on strength of the VTComp question.

Comfortable feature :

In the nutshell, most participants appreciate automatic data extraction from an unstructured text document. Simultaneously, they also desire that extracted data should be exported, which will be very convenient for further practice outside this tool as well.

In the post-study interview, we asked several questions to participants, for getting feed- back about their comfortable(most liked) element in VTComp, we asked them "which was your comfortable feature in VTComp and why?".

The majority of the participants said, the automatically extracted information provided them better guidelines during the creation of data-driven visualization throughout the experiment process, which assist them to create informative visuals easily and within a very quick time. For example, Participant P6 said: " Data extraction from the unstructured document was his favorite feature because manually extraction of data is time-consuming."Similarly, participant P16 also gave a very similar response, he said: "My favorite feature in VTComp is textual information view which helps while creating the visuals by giving us guidelines about the document." P14 said: "I likes the automatic extraction of data and categorizing them like important keywords into separate and easily available position, which are quite useful for annotation purpose during designing. In addition to it, he also mentions about summary and headings sections are also useful because he said that these categories give the overview of the targeted document"

However, we also observed that some of the participants wanted to add HROs which are neither available in extracted data by tool nor in iPad. In that case, the option implemented in VTComp called as online image search as shown in 11’s section b were also preferred by participants and most of the participants said No doubt the option which reduces human efforts is very useful. In addition to it, Participant P16 said: "I likes the online image search in one

platform, otherwise we have to go to search engine, select and download the image, then only we can use it in our visuals which is a quite tedious step, but by providing online image search in one platform reduce the efforts."

Next, we noticed that during the visualization creation process participants frequently used data point adjustment options for the chart. In the interview session, they also notice that data point adjustment and chart generalization without the requirement of coding is a very valuable feature which supports users to choose the chart and data points according to their requirement.

For instance, participant P8 said" He like the chart generalization option which contains chart types according to the behavior of data in the selected chart. The participant further mentions that the user can select the type of chart as per his requirement. Next, he mentions that the ability to reduce the data point inside the chart is also very useful to adjust the visuals according to the visualization target device size."

Overall, we identified the feature "automatically data extraction from unstructured docu- ment" is highly comfortable feature which was responded by 8 participants out of 18, after that online image search option, data point adjustment, annotation toolbar and artboard(5, 3, 3, and 2 respectively) were the most features they user voted as their comfortable tool as shown in Figure23.

Strength of the system:

The strength of the system questions response was up to the marks of our purpose of this work, where participants responded that VTComp is a time saving and easy to use the tool. Several participants stated even less technical person can use this tool comfortably. Some participants also suggested the ability of extraction of data from the unstructured document and creating visuals on it, connect the two different domains( data analytics and visualization) into one platform.

Throughout the task, we observed that participants were excited and also on the same side some were a little nervous. But once they start using the system, they start enjoying it, even the results of the survey also show that participants enjoyed while designing the visualization on VTComp. To observe the participants’ behavior in more depth we analyzed the results of the interview session and their think-aloud logs. For instance, Participant P19 said " I am less technical and was a little nervous before starting the task but after using this system, I can say that this tool is easy to use the tool, even person like me can easily use this". Participant P1 said: “I think less technical journalists can also create data-driven visual confidently because it is a bridge between editor, data-analyst, and designer ". further, participant P16 said:"The strength of this system is the ability to extract the information from document just by URL and the option of chart generalization which helps to modify the charts without the requirement of technical knowledge." One participant P18 describe in a very beautiful way, he said:"Multiple available options under one umbrella added flavors to the user."

Next, we asked these participants to consider reasonable kinds of stories that might serve from VTComp, these involve presenting the peoples’ behavior toward government, comparison

between countries’ affairs, etc. They also suggested different use cases that we were not in our mind, such as using VTComp for data analysis purposes. VTComp can be assisting tools in the domain of education visualization, especially for summary visualization. One participant suggested that it can be an investigative system in statistical reports. Finally, VTComp’s extraction and categorizing the unstructured data can help while preparing for interviews with its key functions like summary, important keywords, and graphical visuals.

Complex feature and Suggestions:

The observation of participants’ behavior when they are unhappy or uneasy with the system can give us very important suggestions. So, during the task, we also notice some of the suggestions by the participants, in which they said that the involvement of exporting extracted data into CSV form or excel would be nice work, which can be further used outside the VTComp environment.

To get proper suggestions and feedback we asked a couple of questions: 1) Which were complex features in VTComp and Why ? 2) What are your suggestions to improve the system? we got a very valuable response which we analyzed and came to know that some people responded that some of the interaction icons are a little confusing. Participant P13 suggested that PDF file data extraction indulgence will be plus point, as currently, this system is only extracting the information from the live document URL.

Besides participants who said extraction from pdf file and exporting, a couple of participants P5 and P19 said: "It would be nice to add more contents inside summary section of textual information view." Furthermore, as currently, our system is supporting only English language documents for processing. So, Participant P8 said: " This system should be enabled to extract the information from multi-language documents and adding the feature to enable animation can also be useful for making more effective and eye-catching visuals"

Overall, after the study, participants acknowledge that VTComp is a very useful tool for creating data-driven visuals in quick time as shown in Figure24. Which can be easily operated by even less-technical persons. Besides, the approach of automatically extracting entities and categorizing from an unstructured text document with just a URL and they no longer have to start with structuring it before using for creation of visual thumbnails. The quick visual response during the authoring process was also highly appreciated. We also took the suggestions during the study and considered some of them as our future work, which includes suggestions like exporting the extracting data into CSV or excel file and providing multi-language data processing support under VTComp.

VI Discussion and Limitations

VTComp proposes a methodology to combine the extraction and format of data from the ar- ticle’s unstructured text, tables, and graphics by using the article URL and designing tools for further updating the model generated visuals. The design of our system’s model is under the assumption that extraction data using NER is decent but not perfect, and the visuals generated by this model can assist to support human-in-the-loop design. However, if the data extraction is perfect, still many sub-tasks need humans involvement such as while designing the interactive informative visualization thumbnails author can add, update or delete components using the artboard provided by VTComp, and that needs nuanced humans judgment often frequently.

In addition to the extraction of the entities using the relation between them, there are other tasks like designing on top of the model generated visuals graphics. For designing interactive visualization thumbnails, the author adds some custom annotations and other storytelling com- ponents that supplement the model produced results. Although in early-stage we designed it by considering the creation of informative visual thumbnails, by the feedback from participants they suggested that this tool can be a supportive tool for data analysis tasks such as changes in subjects by the time. Additionally, partial participants also suggested that this tool can be useful for the presentation of data-driven documents visually and in the domain of educational visualization.

Limitations We have identified several limitations of this system during the experiment. For instance, a couple of participants mentioned that currently, VTComp supports only the English language which is its restriction to process the data-driven news articles from other languages.

we consider this as an opportunity for future work. Next, VTComp limited to produce only static data-driven visual thumbnails means without any animation or other multimedia files. In addition to it, we also observed during the experiment session that many participants suggested that it should also process the data from PDF format as well. Other than this, VTComp takes a little long time approximately (30 seconds) for producing a result that can also be annoying for users. Next, VTComp’s graphical user interface is designed by keeping in mind the iPad model name a1670’s specifications and resolution, however, VTComp can operate from other devices as well because it runs on browser connected with the internet.

VII Future Work

We will seek future opportunities based on VTComp’s current work and experiment results.

While designing VTComp we consider using simplicity over expressiveness, which enables even fewer technical authors to design his/her informative visuals to express the insights of the data using visualization. However, before conducting the experiment we thought, manually inserting or editing data and exporting the extracted data can be a contribution to this tool that was also observed in qualitative study analysis. Next, automatically selecting visualization type by analyzing the extracted data behavior would be a valuable contribution towards automation visualization. Moving further, extraction data points from graph image from the article with the help of computer vision would also be a nice contribution. Further, in our study’s interview session, we got some crucial suggestions like enable data extraction from unstructured PDF format documents as well, which would be a nice step toward common toolkit for different domains. Indulgence of animation would be a big plus for narrative visualization in VTComp.

Additionally, improving extraction accuracy between related entities and reducing process time would make VTComp a more sophisticated tool for designing and exporting intuitive visuals and thumbnails graphics.

VIII Conclusion

Despite the popularity of visualization in both the research and application domain, few re- search studies have focused on the impact of creating interactive data-driven visual thumbnails.

In this study, We presented the authoring tool to support creating interactive visual thumbnails from a document that contains unstructured text, tables, and graphics. It is always a tedious and error-prone process to pick the target entities manually. Its model recognizes temporal, percentage, currency and cardinal expression from unstructured text and tables from the given document’s URL after parsing. It empowers authors to create informative thumbnails from the selected article within a few moments. The selection of designing, annotation and chart compo- nent control elements are decided after closely review the prior work on visual embellishments, visualization memorability, and infographics. Which further modified after performing the pilot study two times to detect bugs, measuring the complexity of the functions and getting feedback from participants.

VTComp allows a wide community of users including those who have a less professional background to operate because it is a browser-based tool that makes easily accessible and its simple interface gives confidence to the author to interact without any difficulties, further to operate VTComp does not require any programming skills. The tools like detection pencil assist authors to find the shape just by drawing it on the artboard surface.

It could be a tool to bridge the gap between journalists, editors, and designers. It can be a useful tool for access to a wide range of potentials authors including journalists, who have periodically deadlines to design informative thumbnails.

Next, Experiment with participants and semi-structured interview results depicts that VT- Comp is a valuable tool not only for data-driven journalism but also useful for other domains like educational and statistics as well. Some participants even point out its use cases for general purpose like for blogger information visuals. Additionally, during the experiment, we find out the system’s strengths and weaknesses like lack of adding animation elements and taking a little long processing time which we consider as future work.

Furthermore, VTComp provides the options to directly export the visuals into the static image format, that further can be used alongside published articles. Interviews and community response provided evidence that the VTComp approach of extraction of the target entities from unstructured text and table, and Scaffolding provides useful and appealing functionality in several application domains.

Acknowledgements

After a little more than two year away from home, I feel extremely grateful to have found the help and support of the wonderful people I met at the Ulsan National Institute of Science and Technology. First, I would like to thank my advisor Prof. Sungahn Ko. I will always be thankful for the dedication, patience and empathy he provided. I really appreciate his dedication to my work and his genuine interest in listening to my ideas in order to give me appropriate feedback.

I would like to thank my committee member Prof. Won-Ki Jeong and Prof. Kwang Soo Kim for giving their valuable time and insightful comments and encouragement. Thanks to the caring staff of the School of Electrical and Computer Engineering, always willing to help me and my lab-mates in every way possible. I also want to thank Professor Kwang Soo Kim, who give me the opportunity to study and learn from such an innovative environment.Further, I thank Staff from Center of international affairs, Ms. Minji Kim, for receiving us with warmth and enthusiasm at the beginning of our first year and for the unconditional support she showed us during our studies. Last, but not least, to my loving family, for encouraging and supporting me during this defining experience.

Dokumen terkait