A paper at WWW 2008, "Spatial Variation in Search Engine Queries" (PDF), by Lars Backstrom, Jon Kleinberg, Ravi Kumar, and Jasmine Novak offered many clever examples of using where people are when they do a web search both to determine when interest in a topic is geographically isolated and to estimate the physical location of objects.
The heart of the matter is this: how do you measure the quality of search results
The first is that we have all been trained to trust Google and click on the first result no matter what. So ranking models that make slight changes in ranking may not produce significant swings in the measured usage data. The second, more interesting, factor is that users don't know what they're missing.
here's the shocker -- these metrics are not very sensitive to new ranking models! When Google tries new ranking models, these metrics sometimes move, sometimes not, and never by much
Two learnings from this story: one, the results depend quite strongly on the test set, which again speaks against machine-learned models. And two, Yahoo and Google users differ quite significantly in the kinds of searches they do
Max Wilson at the University of Southampton recently called my attention to a pair of special issues of Information Processing & Management. The first is on Evaluation of Interactive Information Retrieval Systems; the second is on Evaluating Exploratory Search Systems. Both are available online at ScienceDirect.
Taleb makes a convincing case that most real-world phenomena we care about actually inhabit Extremistan rather than Mediocristan. In these cases, you can make quite a fool of yourself by assuming that the future looks like the past.
The current generation of machine learning algorithms can work well in Mediocristan but not in Extremistan.
It has long been known that Google's search algorithm actually works at 2 levels:
An offline phase that extracts "signals" from a massive web crawl and usage data. An example of such a signal is page rank. These computations need to be done offline because they analyze massive amounts of data and are time-consuming. Because these signals are extracted offline, and not in response to user queries, these signals are necessarily query-independent. You can think of them tags on the documents in the index. There are about 200 such signals.
An online phase, in response to a user query. A subset of documents is identified based on the presence of the user's keywords. Then, these documents are ranked by a very fast algorithm that combines the 200 signals in-memory using a proprietary formula.
This raises a fundamental philosophical question. If Google is unwilling to trust machine-learned models for ranking search results, can we ever trust such models for more critical things, such as flying an airplane, driving a car, or algorithmic stock market trading? All machine learning models assume that the situations they encounter in use will be similar to their training data. This, however, exposes them to the well-known problem of induction in logic.
My hunch is that humans have evolved to use decision-making methods that are less likely blow up on unforeseen events (although not always, as the mortgage crisis shows)
A newly published patent application from Microsoft takes an interesting spin on presenting information, pulling together news from a mix of sources to present topics in storylines, and providing ways to have that information delivered to us over computers, smart phones, watch interfaces, and in other ways.
The standard measure scientists use to judge the importance of scientific papers is a simple citation count. That is, how many other papers cite the paper in question? While this measure has its merits, it has one fundamental flaw - not all citations are equal.
Numerous authors/bloggers have advocated using a PageRank-like index for quantifying the importance of papers or journals
To represent the web we use a directed graph, where the edges carry a direction.
The goal of the PageRank algorithm is two-fold. We wish to construct a measure of relevance that, first, is related to how many incoming links a site has, and second, what the importance of the source of those links was.
Well scientific papers can be mapped to a graph in a similar way to web-sites. Specifically, vertices in the graph would represent papers, and edges citations. The PageRank algorithm can be applied out-of-the-box.
First of all, one could discount self-citations from the index
A second variation that one might try is to add a time bias when calculating the index, such that links from more recent papers carry more weight than from older papers.
刚刚结束的 WWW2008 会议里有一篇 short paper,《Size Matters: Word Count as a Measure of Quality on Wikipedia》。里面给出了一个令人吃惊的实验结果,在进行 Wikipedia 的文章质量评价时,仅仅只需要使用"Word Count"一个参数,就可以取得 96.31% 的准确率!这个结果,比许多使用复杂模型的算法,都要好!
A new index measuring a scientist's impact in his/her field has been developed called the Wu index or w-index. Developed by Qiang Wu from the University of Science and Technology of China in Hefei, it was published as The w-index: A significant improvement of the h-index in this week's Physics arxiv.
Sixty years ago, digital computers made information readable. Twenty years ago, the Internet made it reachable. Ten years ago, the first search engine crawlers made it a single database.
Google's founding philosophy is that we don't know why this page is better than that one: If the statistics of incoming links say it is, that's good enough.
The scientific method is built around testable hypotheses. These models, for the most part, are systems visualized in the minds of scientists. The models are then tested, and experiments confirm or falsify theoretical models of how the world works. This is the way science has worked for hundreds of years.
Peter Norvig, Google's research director, offered an update to George Box's maxim: "All models are wrong, and increasingly you can succeed without them."
Once you have a model, you can connect the data sets with confidence. Data without a model is just noise.
There is now a better way. Petabytes allow us to say: "Correlation is enough." We can stop looking for models. We can analyze the data without hypotheses about what it might show. We can throw the numbers into the biggest computing clusters the world has ever seen and let statistical algorithms find patterns where science cannot.