My first year as a Research Postgraduate student

It’s mainly a matter of changing my mind. I am no longer a UG student that I had a PG mentor to work with. I have to find my own research topic, and proceed it by myself. As there is a big change on what I am expected to do, my perspectives toward research needs to change too.

Which research idea is good enough (i.e not mediocre) to write a paper?

For sure, in order to be an independent researcher, that’s what I need to learn the first. A year ago, as a naive new-comer, I have an expectation (I believe that it’s similar to that of other new-comers too) that my research paper need to be ‘fancy’, ‘valuable’, a very strong paper! It should be a methodology-focus paper, because it means I can solve a problem. I am good. Other types of research, such as introducing datasets, survey, etc. are not of my preference.

However, ya, a year later which is now, I wake up from that dream and be realistic. In the last one year, I mainly worked on 3 main projects, and 2 side projects. 2 side projects are ChatGPT evaluation and a second-author paper on causal commonsense reasoning. I learned from these two projects that the methodology is not so important. The former, even though being a position paper which just did evaluation and discussed the result, really met what people needed at the time, as ChatGPT had been released two months and there was no empirical study on that Language Model. Meanwhile, the later proposed a new problem (not really new, as there is an anchor previous work), and introduced a new dataset. It is full of theory, and not much about methodology. Yet it’s accepted to ACL Main Conference. Mind blowing! It seems that what I think is not “good enough” is actually (very) good to publish a paper.

How about the main projects that I work on? Learning via doing is much more mind-imprinting, right? Yep, I pursued a research about conceptualization in the last Fall semester. It is the subsequent work of two paper, my first paper in which I am the second author and another paper in my research group. If I can successfully apply “conceptualization” to the task CSKB Population, that would be a great contribution to write a paper and a great contribution to the overall research of my lab. That would impress my supervisor, and make me a strong candidate in the upcoming PhD application period. But, as usual, thing is not that easy. I spent the whole semester with that project, try a lot of stuff, but the result remained unchanged - no improvement yet! When I came back to this project later (and because I used another approach, I count it as a separate project), I still dreamed of using conceptualization capability of LLMs to aid CSKB Population task. At the moment I once again failed, my supervisor said to me: “Quyet, you don’t need to solve the task using conceptualization. What you did in the error analysis actually guides you what can be improved. You can try any method, as long as it works then it’s good”. That advice meant a lot to me. I understood that we all dream for beautiful stories and extraordinary methods for our papers, but research is often just incremental, ad-hoc like that. From that moment, I accepted that my first first-author paper’s method can be mediocre. I iterated quickly, digged into analyzing errors of previous methods, and came with a solution. After 6 weeks from the meeting with my supervisor, I finished the paper and submitted to EMNLP conference. What an experience for me!

I know that in the future, when I am a senior researcher, I may be “not allowed” to do such a kind of work anymore as it’s mediocre. However, at this moment, I am still a junior rpg student. There are many things for me to learn, and my supervisor doesn’t expect me to produce a great work yet, as “Tuổi nhỏ làm việc nhỏ, tùy theo sức của mình”. Nonetheless, later when I need to produce great work, I may still follow this research approach - start with small problems first. As a famous quote in business is, “Think big, add small, move fast”!

What kind of research I can/should do in my first year?

During the Research Grant Council (RGC) Visit to HKUST in this June, I had a chat with a Professor from University of Bath. He already retired, but join the tour to meet students around the world. He shared to us his research experience, and his supervision experience too. What I remember the most is about what kind of research should we do in the early stage of our research career. There are 5 types of research (as far as I know), which are theory, methodology, application, position, survey. As junior research students who lack knowledge on our own research fields, we can start with a survey paper. “Find a topic that there is no survey yet, then write a survey on the topic. It’s surely a contribution to the research community, as well as an opportunity for us to study our research field thoroughly”. I think it’s a good research path to follow. To be honest, there are a huge horizon left for me to explore in the field of NLP. I am still far from mature in this field. Doing a survey paper can be a good chance for me to grasp more knowledge on the field and acquaint myself with popular NLP tasks and methodologies. After I am well-equiped with knowledge, it’s easier for me to do other types of research. When I start my PhD, I will definitely follow this research path. In the future, if I become a professsor, I will also encourage my students to follow that research path.

My perspectives toward research has changed a lot after 1 year. I believe that the process still continue to complete me.