Paper Review: Finding Question-Answer Pairs from Online Forums

This paper was written by Cong (Aalborg University) et al. and presented at the 31st annual international ACM SIGIR conference on Research and development in information retrieval, 2008.

Question-Answer System is currently a very hot topic in the IR community and attracted many researchers. This 2004 paper (published in ACM SIGIR’08) is one among many in this area. The problem the paper tries to solve is how to mine knowledge in the form of question-answer pairs specifically from a forum setting. The knowledge can then be used for QA services, to improve forum management, or to augment the knowledge base of chatbot. This is a challenging paper to read because it touches many different concepts and ideas from many research disciplines besides IR such as machine learning (n-fold cross-validation), Information Theory (Entropy), NLP (KL-divergence), Bayesian statistics (Markov Chain Convergence), and Graph Theory (graph propagation).

The main contributions of this reference paper include: 1) a classification-based method for question detection by using sequential pattern features automatically extracted from both questions and non-questions in forums; 2) an unsupervised graph-based propagation approach, which can also be integrated with classification method when training data is available, for ranking candidate answers. The paper also presented a good amount of experimental results including a 2 (datasets) 4 (methods) 3 (measures) design for question detection, a 2 (w. w/o answers) 9 (methods) 3 (measures) design for answer detection, and a 2 (w. w/o answers) 2 (KL convergence or all three) 2 (propagate w. w/o initial score) 3 (measures) design for evaluating the graph-based method, and showed the performance superiority of the proposed algorithm compared to existing methods.

The paper proposed several novel ideas in solving the proposed problem. First, the paper defined support and confidence and also introduced minimum thresholds for both, which are additional constraints previous works did not have. “Minimum support threshold ensures that the discovered patters are general and minimum confidence threshold ensures that all discovered LSPs are discriminating and are cable of predicting question or non-question sentences.” (Quoted from the paper.)

Second, the paper introduced the idea of combining the distance between a candidate answer and the question, and the authority of an author, with KL convergence using a linear interpolation. Because of the specific forum setting, these additional factors improved the algorithm performance. However, note that these additional factors also poses limit to the applicability of the algorithm to other Question-Answer mining applications.

Third, the paper proposed a graph based propagation method that uses a graph to represent inter-relationships among nodes (answers) using generator and offspring ideas to generate edges (weights). With this graph, the paper suggests propagating authority through the graph. The authors argued (briefly) that because this can be treated as a Markov Chain, therefore, the propagation will converge. This idea of using a graph to propagate authority information is great because it takes into consideration of how inter-relationship between pair of nodes can be used to help ranking (following the PageRank idea). The idea of integrating classification (two ways) with graph propagation is another great idea. However, I find this Markovian argument weak. No rationality is given about why this can be treated as a Markovian process, and the transitional probability mentioned is not convincing.

The experiment design in this paper is extremely good. First, when using two annotators to annotate the dataset, the paper created two datasets, the Q-Tunion and Q-TInter and evaluated different algorithms using both datasets. This effectively shows that the algorithms performances showed same trends even with disagreeing annotators. The paper also showed detailed performance comparisons using multiple measures across different datasets and different algorithms/methods. This way, the superiority of the proposed algorithm is clear and convincing.

Additionally, the paper used many good examples in the first half of the paper explain complex concepts or to provide justifications. This is good writing! I wish the paper had done the same thing for the latter part of the algorithm description.

The paper also has some significant drawbacks. First, the paper certainly covered a great deal of information and ideas, especially because of the experiment design, a large amount of performance values are contrasted and analyzed. Even the authors used the phrase “due to space limitations” three times in the paper. It is truly a difficult task to cram everything into 8 pages, which is a constraint for a conference paper. And by doing so, lots of useful information are omitted (e.g. the section about how the Markov process can be justified) and some part of the paper just seemed difficult to understand (section 4.2.2). There are also places where proper references should be given, but are omitted probably due to space limitations (e.g. references for Ripper classification algorithm and power method algorithm). It is probably a better idea to either publish this as a journal paper where more space is available, or write a technical report on this subject and reference to it from this paper.

Second, it also seems that the authors were in a rush to meet the paper deadline. This is shown by the carelessness in many of the math notations used in the paper. For example, the Greek letter λ is used in equations (3), (5), (6), and (10) where they meant different things. The letter ‘a’ used to represent an answer sometimes is bolded and sometimes is italicized when all of them meant the same thing. Recursive updates are also represented using “=” instead of “”, such as in equation (10), and temporal indexes are not used. There are quite a few important spelling errors such as the sentence right after equation (3) where ‘w’ was written as “x”. My personal opinion is that if you are going to put your name on the paper, then better show some professionalism.

Third, the paper proposed a quite complex model with many parameters. Especially, the algorithm used many parameters that were set by empirical results. The author did mention that they did not evaluate different parameter values in one place and discussed the sensitivity of the empirical parameter in another; however, these empirical parameters make one wonder whether the algorithm will generalize well with other datasets or whether these parameters might be correlated in someway. A better approach would probably be either justify qualitatively or quantitively the sensitivity of these empirical parameters by discussing more intuitions behind them or showing experiment results using different values with several datasets of different domains and scale. The paper also proposed a few “magical” equations, such as author(i), equation (10), without rationalize how the formulas came about. (Again, publishing as a journal paper would have lessened such problems.)

There are other minor problems with the paper as well. For example, the paper mentioned several times that the improvements are statistically significant (p-value < 0.001), but without much more detail on how the statistical significances are calculated, I can only assume that they came from the 10-fold cross validation. In my opinion, statistical significance would not be a very good indicator of improvements in such set up. The paper also gave me the impression that they are ranking candidate answers by P(q|a). I would think ranking by P(a|q) would have been more appropriate.

Video of the Day:

Now here are some serious question asking and answering!

Full Moon Crescent Saber: Chapter 1 (1)

Chapter 1: The Outstanding Youngster

It was dawn, foggy, very thick fog.

Ding Peng pushed open the window of his small room. Milky, thick fog drifted in like willow catkins and stroked his face gently.

He had a very delicate face, and a healthy body. When he spoke, he looked vigorous, full of vitality. Yet when he smiled, his face often showed a childish and naïve look, as though he was just a big boy, a kid you had always watched as he grew.

But Ding Peng was no longer a kid.

In the last three months, he had already defeated in a row three famed swordsmen in the Martial World.

Sunshine and water make plants and trees grow bigger and stronger. Victory and success also make a boy mature and grow up.

Now he was not only a true man, but also calm, composed, and full of confidence.

He was born in March, and turned twenty this year. It was right on his birthday when he defeated the renowned swordsman Shi Ding of the Baoding City, using a move called “Shooting Star Beyond Skies.” Shi Ding was a sword master of the Northern Style Green Duckweed Sword Art. He used this victory as a birthday present for himself.

In April, he defeated “Wind-Chasing Sword” Ge Qi with the same move “Shooting Star Beyond Skies.” Ge Qi was the head apprentice of the Huashan Sword School. His sword art style was especially speedy and peculiar, and contained many ruthless moves.

Ge Qi was a very proud swordsman, but after that fight, he admitted his defeat sincerely. He actually acknowledged publicly, “Even if I train for another ten years, I would still have no way of fending off that sword move of his.”

In May, the Head Master of the Iron Sword Clan, “Songyang Swordsman” Guo Zhengping, also lost to his sword move “Shooting Star Beyond Skies.”

Guo Zhengping’s comment for this sword move and for him was, “The sword move rises above the mundane world, leaving absolutely no trace behind. Within one year, this young man will shine in the Martial World and have the world under his feet.”

The Iron Sword Clan was not a prominent school, but it had a long history and a reputation of honesty and decency. Naturally words from the Head Master of such a Martial Art school carried extra weight.

Every time when Ding Peng remembered that comment, he couldn’t help but feeling a great surge of excitement and thrill.

“Shine in the Martial World, and have the world under his feet!”

He had trained very hard in the past thirteen years, fourteen hours every single day, so hard that both his palms and soles bled many times from excessive rubbing.

Especially during those coldest, bitter nights of the winter, to maintain an upbeat spirit, he always held a snowball in his hand. If he saw any sign of sluggishness in himself, he would shove the snowball into his pants, a kind of experience not easily imaginable by others.

The reason why he tortured himself like that was because he was determined to excel and succeed, to earn respect for his father who accomplished nothing in his entire life.

His father was a nameless escort, and accidently got hold of an incomplete page of a sword art manuscript.

The entire sword art only contained one page. And on that one page, there was only one move, “Shooting Star Beyond Skies.”

A shooting star flying by from beyond skies and suddenly disappears. The instant of shine and swiftness are simply unstoppable. But his father was already too old with declined intellect and sluggish response. He could no longer master such speedy sword art. So he passed the page of sword art manuscript to his son.

The last words he left his son at his death bed were, “You must master this sword move. You must earn respect for your father so everyone knows that I have a son who has the world under his feet.”

Every time when Ding Peng thought of his father’s words, his blood would boil and he couldn’t hold his tears back.

But now he decided to never shed another drop of tear. Tears are for the weak. A true man only shed blood!

He took a deep breath of the early morning’s fresh air and drew his sword from under his pillow. Today he would once again fight for another victory using this same sword move.

If he wins today, it would be a real success.

Though Shi Ding, Ge Qi, and Guo Zhengping were also famed swordsmen in the Martial World, the three previous victories would be no match for today’s battle.

Because his rival today is Liu Ruosong.

A member of the “Three Friends of Winter^[1]”, the world-renowned “Pine Swordsman”, Liu Ruosong; the owner of the “Ten-Thousand Pines Villa”, Liu Ruosong; the only layman apprentice of the prominent Priest Tianyi of the Xuanzhen Taoist Temple on Mount Wudang, Liu Ruosong.

Many years ago had he heard of this name. At that stage of his life, this name had been as grave as Mount Taishan^[2], majestic, lofty, and untouchable. But that was no longer the case. Now he was confident that he could defeat this man. He followed the most appropriate manner when he sent in his invitation, requesting an opportunity to learn swordsmanship from an eminent senior sword master, to make sure Liu Ruosong could not refuse the fight, because he must defeat this man in order to advance and join the class of true Kung Fu masters in the Martial World. Both the time and place of the challenge were set by Liu Ruosong – June 15^th, noon, at the Ten-Thousand Pines Villa.

Today is June 15^th.

The fight today will shape his fortune for the rest of his life.

The clothes he had washed, straightened, and put up on bamboo poles by the window last night were almost dry now.

Although they were not completely dry, they would soon be once he put them on.

This was the only set of clothes he owned, hand-sewed by his old and sick mother right before his departure. It had been washed pale by now and had many frays. But as long as it was washed clean, he found no shame wearing it.

Poverty is not shameful; Laziness and dirtiness are.

He put on the clothes, and then retrieved from underneath the pillow a money bag also made of blue cloth.

There was only a small knob of silver inside.

This was his entire asset. After paying for the small inn room, there would probably be only a few dozens of copper pennies left.

Normally he slept at places that required no rent. Under the sacrifice altar in the shrine or on the small meadow inside the woods could all be his bed.

For the fight today, he gritted his teeth and booked a room at the small inn, because he needed plenty of sleep to make sure he had plenty of mental and physical strength to win the fight.

After taking care of the inn bill, he made the harsh choice of spending the rest of the money on half pound of halogen beef, ten pieces of dry bean curd, a big bag of peanuts, and five big steamed buns.

For him, this was not only extreme luxury, but also unforgivable waste. Usually he could easily get by a day with only three hard bread cakes.

But he decided to forgive himself today. Today he needed strength, and strength only comes with quality food.

Besides, things could be completely different after today.

Fame can bring not only glory and self-esteem, but also many things one would never dream of. Wealth and social status will also follow along.

He knew that very well, hence he always gritted his teeth to endure poverty and hunger.

He never let himself be tarnished by any indecency, for he was determined to rise up through the right way.

Since it was still four hours from noon, he decided to find a good spot to enjoy the wonderful food.

Somewhere by the foothills near the “Ten-Thousand Pines Villa”, he successfully found a spot that had a spring, a good size patch of grass, beautiful flowers and a nice view. The place was surrounded by flowery trees. The cloudless sky radiated with a beautiful shape of blue.

By then the thick fog had all cleared. The sun had just barely risen. Morning dews glittered on top of dark green leaves like beautiful pearls.

Ding Peng sat down on the soft grass and tore off a piece of beef. The beef tasted even better than what he had imagined.

He was extremely delighted.

Just at that moment, a girl ran into his secret little world like a little antelope chased by a hunter.

The girl was completely naked.

---

[1] “Three Friends of Winter” was created by Song Dynasty Poet Su Shi and refers to Pine, Bamboo, and Calyx canthus in Chinese culture because all three of them remain beautiful despite the severe cold of the winter. Here it is the name of a small trio of Kung Fu Masters whose nicknames contain these three things. The last character in Liu Ruosong’s name, “song”, also means pine.

[2] Mount Taishan is a famous mountain in Shandong Province, China. It is the first of the “Five Mountains” and is known for its majesty.

Now support the translator Lanny by following my blog and leaving comments! :)

Do not pretend to be something you are not, and I am not a plumber!

AI and Robots: Hybrid Video Display of Visible and Infrared Might Help Search and Rescue

A recent article in New Scientist discussed a research project performed at Brigham Young University where the effect of combining visible light and infrared on real-time video footage in Wilderness Search and Rescue was evaluated.

The research project was directed by Dr. Bryan Morse (who is also one of my committee members) and implemented by Nathan Rasmussen (a friend of mine, who successfully received his MS from this project and graduated in 2009). It is one of the many projects in the WiSAR research group at BYU that works on how to use mini-UAVs (Unmanned Aerial Vehicles) to support Wilderness Search and Rescue. The picture on the right shows Nathan throw-launching a UAV in a field trial at Elberta, Utah.

This research focuses on the human-robot interaction aspect and try to determine which method of display works better for human operators: displaying visible light video side by side with infrared video, or combine both in a hybrid display.

The UAV used in the experiments can already carry both a visual spectrum camera and an infrared camera (BTW: very expensive). Visible light video footage can be useful in spotting objects of unnatural/irregular shapes and colors (top view). Infrared light video footage, on the other hand, can be helpful in detecting objects with distinct heat signatures that are different from surrounding environments (especially early mornings, evenings, and nights, or in cold weathers where heat signatures are more distinct).

In order to align footage from both sensors, a calibration grid was created with black wires on a white background. To allow the infrared camera to "see" the grid, an electricity current was sent down the wires to heat them up. An algorithm is then used to align the vertices of the two grids to compensate for the slightly different viewing angle.
Once the hybrid view was possible, a user study was performed where students were used as test subjects to watch UAV videos in both methods and tried to identify suspicious objects while listening to audio signals (counting beeping sounds as a secondary task in order to measure mental workload). I happen to be one of the test subjects, and my hard work earned me some delicious chocolates.

Experiment results show that people who viewed the hybrid display performed much better in the secondary task of counting beeps. This suggests that the hybrid video is easier to interpret (requiring less mental work) and would allow the searcher to focus more on identifying objects from the fast moving video stream.

The research was presented at the Applications of Computer Vision conference in Snowbird, Utah, in December 2009. If you are interested in more details about this research, you can read Nathan's thesis　(warning: 22.4MB).

Picture of the Day:

Beautiful dusk sunshine mountain view from my house!

Robot of the Day: CubeStormer, the Rubik's Cube Solver

Ever played Rubik's Cube before? If you have, then you know how hard and how long it takes to solve a game of Rubik's Cube. The robot we talk about today, however, can solve it within seconds, and most impressively, the robot was built completely using Lego pieces from the Lego Mindstorms Kit, which means you could build a robot just like this yourself for only a couple hundred bucks!

The robot's name is CubeStormer, built by British engineer Mike Dobson using Lego Mindstorms parts hooked up to a laptop computer. The computer acts as the brain and performs tasks such as recognizing colors, solving the puzzle using algorithms, and sending motor commands.

As shown in the video below, the robot first quickly inspects all six sides of the cube using multiple cameras by first rotating it a few times to recognize the current state of the cube. The computer vision task is actually really simple because the cube is placed at a fixed position, so the recognition software only needs to sample a few points for each color piece and then simply detect the color of the pixels (one out of six possible choices). The state of the cube is then passed on to a solver software (such as this free online one) and sequences of moves are generated, which are translated into motor commands for the robot to perform.

What is impressive about this robot, though, is the engineering side of things, such as how parts are connected and how motors are used all with toy Lego pieces. A beautiful designed enabled the robot actuators to solve the game in such short period of time. If you look closely at the video, you'll also notice that two rows of the cube can be rotated at the same time to speed it up!

CubeStormer by Mike Dobson

The time it took the robot to solve a random game was about 12 seconds. This is very much comparable to the fastest human Rubik's Cube solvers such as the one shown below.

Rubik's cube official world record 7.08 Erik Akkersdijk

There are of course other Rubik's Cube solving robots in the wild, such as the one built by UC Berkeley shown in the video below, which solved a puzzle in 6 seconds. But apparently this robot would cost a lot more.

Rubik's cube solver by UC Berkeley

However, the Cubinator, aka RuBot II, by Pete Redmond from Dublin, Ireland gets extra point in my book of Human-Robot Interaction. Although much slower compared to the other two robots, it has a head and two arms. And after picking up the cube all by itself, it even played music and talked to the audience while solving the puzzle.

Cubinator by Pete Redmond

What if the Cubinator not only solves Rubik's Cube, but is also capable of playing board games or hide-and-seek with your kids, tell them jokes, read books for them, and help them with their homework? Would you want one for your kids? If so, for how much? If not, why?

Picture of the Day:

Leftover Valentine’s chocolate? Use it to measure the speed of light with your microwave. Click the picture to find out how!

Random Thoughts: Are you being watched?

It started as a great idea at the Lower Merion School District, outside Philadelphia, when school officials decided to loan laptop computers to students to encourage them to embrace newer technologies and to study better. However, the situation came to a dramatic turn when a suit was filed against the school district for spying students at students' homes through the web cam built into the laptop and a software that allowed the school officials to activate the camera remotely in order to view and take pictures of the students.

The interesting part was that a student named Blake Robbins was accused of selling drugs and taking pills by school officials (Blake claimed that he was eating candies), and the school official actually provided proof -- images of Blake eating things at home taken secretly through the web cam -- to back up their claim! The image on the left shows Blake's family and their lawyer appearing on "Early Show Saturday Edition" discussing the entire fiasco (photo credit CBS).

I am simply AMAZED at the intellectual capabilities of the school officials involved here!! The issue here is not whether the kid took pills or not. The issue is about a crude invasion of personal privacy at people's own homes without their knowledge and consent. A federal judge quickly ordered the school district to stop activating the cameras, and the school complied. The FBI has also opened a criminal investigation of the web cam use to see if the school district broke any federal wiretap or computer-intrusion laws.

Disturbing as the story is, what I want to emphasize here, though, is slightly different. Many people use web cams to do video conferences with friends and family. Some still use the old external web cams connected through USB ports (I am one of them). However, most people use the laptop built-in web cams these days because of the convenience and also because most laptops come with built-in web cams. So do you really know if you are being watched?

People these days take their laptops with them everywhere they go, including very private places such as their bedrooms and bathrooms. Many companies would also give their employees laptops so they can work from home or while they go on trips. Many large organizations DO put remote access software on company laptops for the ease of IT support, and I personally have used such software when I worked as an IT support staff in the past. The truth is, IT support staff can remotely watch your monitor screen when you have no idea they are doing so. And if the computer has a web cam, activating the web cam through such management tools is a piece of cake.

Also, many home computers are hacked and made into "zombie" computers for spamming or Denial Of Service attacks in large botnets controlled by hackers. These hackers can also easily control the web cams connected to these infected computers and the users would have no clue about such activities! If you think you are safe behind security firewalls you purchased from McAfee of Symantec, you'd better think twice.

So what lesson should we learn from this? That is, we should never naively believe that we have total control of web cams connected to our computers. The safest thing to do is to cover it up with tape of a piece of paper when we are not actively using it, because you never know who might be watching through the web cam. Especially for people have built-in web cams, it is so easy to forget about that special "eye" in the room, and it might be watching you actively at this very moment!

The commercial in the YouTube video below might seem funny, but it wouldn't be so funny if you weren't in a video conference yet someone is secretly watching you behind the web cam device. I think all built-in web cams should have a cover, so people only open it when they actively use it and can always close the cover when they are not.

This also poses an interesting question about future robots in people's homes. These robots will probably also have eyes, and eyelids will probably have to be mandatory so they don't peep on you when you don't want them to.

Video of the Day:

Sony laptop with built-in web cam ad!