2020 Fiscal Year Research-status Report
Measuring Group Interaction in Online Discussions and Application to Autonomous Agent Deliberation
| Project/Area Number |
20K11936
|
|---|
| Research Institution | Kyoto University |
|---|
Principal Investigator |
Rafik Hadfi 京都大学, 情報学研究科, 特定助教 (30867495)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
伊藤 孝行 京都大学, 情報学研究科, 教授 (50333555)
|
|---|
| Project Period (FY) |
2020-04-01 – 2023-03-31
|
| Keywords | Group Interaction / Similarity Metrics / Collaborative Editing / Time Series Analysis / Automatic Deliberation / Entropy Methods |
|---|
| Outline of Annual Research Achievements |
The results in the first fiscal year revolve around designing and implementing new metrics to quantify the interactions between multiple stakeholders working on producing textual content. The developed metrics were tested on collaborative editing on Wikipedia, and were recently published with collaborators from Kyoto University as “Hui Chen Chou, Rafik Hadfi, Donghui Lin and Takayuki Ito. Identifying Collaborative Editing Traits and Phases in Good Wikipedia Articles. The 9th ACM Collective Intelligence 2021". Other relevant results are finalised and are expected to be published as "Measuring Interaction in Collaborative Editing on Wikipedia” in the 18th International Conference on Distributed Computing and Artificial Intelligence, and as "Category Theoretic Approach for Case-Based Reasoning in Hierarchical Problem Domains" in the 29th International Conference on Case-Based Reasoning. The framework proposed in the second contribution is based on information theoretic similarity metrics applied to discussion hierarchies. This outcome contributes considerably to the first objective of the project. Another relevant result is a new entropy-based method to quantify the readability of online discussions from structural complexity. Finally, a recent result is expected to be published as one survey on the concept of deliberation in social and political sciences. The goal from this survey is to explore possible ways to mathematically define deliberation and to implement it algorithmically in artificial agents.
|
| Current Status of Research Progress |
Current Status of Research Progress
3: Progress in research has been slightly delayed.
Reason
The delay was mainly caused by the PI and Co-I changing their affiliations in October and December 2021, respectively, from Nagoya Institute of Technology to Kyoto University. This resulted in logistic and organizational restructuring in their teams and collaborators. In addition, this change of affiliation at the time of COVID was followed by slow administrative procedures to re-allocate the funds from one institution to the other. These delays were amplified by the COVID situation, confinement, social distancing, and the cancellation of several symposiums, research meetings, as well as planned social experiments.
Despite the slight delay, there is substantial progress in the implementation of the research itself (mathematical modeling, algorithms, coding, data generation and analysis). I am therefore expecting to publish most of these findings in international conferences and journals within the upcoming 2 to 6 months.
|
| Strategy for Future Research Activity |
The future plan is to first publish all the proposed interaction metrics for two textual domains: collaborative editing on Wikipedia, and online discussions on D-Agree. Second, the proposed metrics will be algorithmically integrated in artificial agents that autonomously analyze the textual content, and them improve the deliberation process by optimizing over the metrics. Furthermore, such agents will be deployed in online discussions on the D-Agree platform with humans and have their performance evaluated. Herein, performance is defined with regard to the quality of the content, and the degree of interaction between the humans. The interaction will be quantified using the metrics developed in objective 1. The quality will be inspired from the quality metrics used in Wikipedia article generation, and which were encountered when analysing collaborative editing data. There are no important changes in the original plan and in the proposed methodology. The use of Wikipedia collaborative editing data turned out to be beneficial for the evaluation of the metrics at a time where organising social experiments with human subjects was difficult. In the next phase, we will deploy the metrics and the agents in online discussions to assess the deliberative capabilities of the agents. The survey on deliberation will identify the important deliberative criteria, which will serve to benchmark the agents' capabilities.
|
| Causes of Carryover |
Because the PI did not have the chance to use the amount last year due to the slow administrative procedures to re-allocate the fund from the previous institution of the PI (Nagoya Institute of Technology) to his current institution (Kyoto University). These procedures were slowed down by the time the PI changed his affiliation (end of year, December 2021) and by the COVID situation. The usage plan is to allocate the amount as detailed in the application, for journal publication fees, registration in conferences, and for the purchase of books and a computer.
|
Research Products
(1 results)
