Automatic Summarization in Medium
Trying out how TextTeaser will perform
This is part 1 of the series of posts I made about automatic summarization and their application to popular websites.Here’s the next part.
So I have this algorithm. I call it TextTeaser. I formulated this algorithm for my MS research.
TextTeaser is an automatic summarization algorithm. There are two kinds of automatic summarization: abstraction, and extraction. Abstraction is hard. The algorithm do extraction. During my research, I tested TextTeaser using news articles with other related systems. It performs well. My news reader Readborg is currently showing what the algorithm can do.
Then I saw Medium. Generally, Medium is not a news site. I don’t know how my algorithm will perform here. I’m curious so I tested the top posts in Medium. So here they are:
- I recommend McDonald’s
- Everyone unanimously agrees that we can’t possibly go to McDonald’s, and better lunch suggestions emerge.
- I call it the McDonald’s Theory: people are inspired to come up with good ideas to ward off bad ones.
- Lamott, Nike, and McDonald’s Theory are all saying that the first step isn’t as hard as we make it out to be.
- McDonald’s Theory teaches us that it will trigger the group into action.
- I will no longer encourage aspiring artists to attend art school.
- Unless you’re given a full ride scholarship (or have parents with money to burn), attending art school is a waste of your money.
- I have a diploma from the best public art school in the nation.
- Prior to that I attended the best private art school in the nation.
- That any art school should deceive its students into believing that this is a smart decision is cruel and unusual.
- Apple rejected it.
- My latest rejection, however, has me fuming — of a weather app I devloped a few months ago.
- Now, I personally see weather apps, like to-do apps and flashlight apps before them, to be one of those lowest-common-denominator apps.
- You might understand my shock when they unveiled a revamped weather app today.
- According to Apple, no one wanted a flashy weather app.
- It’s been five months since the New York Times dropped their mammoth digital story “Snow Fall,” and some people still talk about it as if it came out last week.
- At a conference recently, the editor-in-chief of the Times said that “Snow Fall” has become a verb inside the newsroom.
- It’s an unreasonable and baffling request for the New York Times to tell us to take down this statement.
- The backlash to “Snow Fall” is that it’s an indulgence only the Times can afford.
- The New York Times responded to my question: Dear Mr. Brown: We are offended by the fact that you are promoting your tool, as a way to quickly replicate copyright-protected content owned by The New York Times Company.
- Since publishing ABC: Always Be Coding - How to Land an Engineering Job, many have asked how I got an engineering job at Google without a college degree.
- Because I didn’t have a formal CS degree, I knew I lacked a lot of fundamental knowledge.
- I was literally obsessed and wouldn’t stop until my fear of the Google interview turned into confidence and excitement.
- I remember every single one of my interviews at Google and had a blast with all of them.
- After an amazing five years of learning and growing as an engineer, I no longer work at Google.
I also tried the algorithm in non-english article. TextTeaser is language independent but I don’t know how good it performs in this article:
- Não há uma carta de reinvindicações na mesa que possa ser negociada.
- Isso é ótimo por alguns efeitos que os protestos deixarão nos governantes.
- Em menos de uma semana surgiu do nada, sem uma liderança, sem aviso ou previsão um movimento que levou centenas de milhares às ruas.
- Há uma explicação para tudo isso.
- Apesar de toda a anarquia, há uma lógica por trás do que estamos vivendo.
If anyone that can understand Portuguese, can you tell me how was it.
There. Those are the most important sentences as obtained by TextTeaser. The purpose of the algorithm is not to replace the original article. It aims to provide a gist of what the article is all about. It’s like the summary at the back of a book, it teases you to read it. That’s why I call them “teasers”. It is also like the movie teasers, it doesn’t show what the movie is all about but it can be enough for you to watch it.
Who says only movies can have a teaser?
So what if Medium’s homepage have these teasers? Well, I tried modifying it and this is the result.
Imagine seeing those teasers in Medium’s homepage. Imagine the authors reading those and teasing you to read their article. It sounds nice.
Update: API is now on Mashape.