short text syntactic classification - solr

I am newbie at machine learning and data mining. Here's the problem: I have one input variable currently which is a small text comprises of non-standard nouns and want to classify in target category. I have about 40% of total training data from entire dataset. Rest 60% we would like to classify as accurately as possible. Followings are some input variables across multiple observations those are assigned 'LEAD_GENERATION_REPRESENTATIVE' title.
"Business Development Representative MFG"
"Business Development Director Retail-KK"
"Branch Staff"
"Account Development Rep"
"New Business Rep"
"Hong Kong Cloud"
"Lead Gen, New Business Development"
"Strategic Alliances EMEA"
"ENG-BDE"
I think above give idea what I mean by non-standard nouns. I can see here few tokens that are meaningful like 'development','lead','rep' Others seems random without any semantic but they may be appearing multiple times in data. Another thing is some tokens like 'rep','account' can appear for multiple category. I think that will make weighting/similarity a challenging task.
My first question is "is it worth automating this kind of classification?"
Second : "is it a good problem to learn machine learning classification?". There are only 30k such entries and handful of target categories. I can find someone to manually do that which will also be more accurate.
here's my take on this problem so far:
Full-text engine: like solr to build index and query rules that draws matches based on tokens - word, phrase, synonyms, acronyms, descriptions. I can get someone to define detail taxonomy for each category. Use boosting, use pluggable scoring lib
Machine learning:
Naive Bayes classification
Decision tree
SVM
I have tried out Solr for this with revers lookup though since I don't have taxonomy available at moment. It seems like I can get about 80% true positives (I'll have to dig more into confusion matrix to reduce false positives). My query is bunch of booleans terms and phrases with proximity and boosts; negations to reduce errors. I'm afraid this approach may lead to overfit and wont scale.
I am aware that people usually tries multiple modeling techniques to achieve which one works best or derives combination of techniques. I want to understand this problem with feasibility and complexity point of view. If its too broad question please just comment on feasibility of solution.

Related

Automatic product classification and query weighting

I'm facing ranking problems using solr and I'm stucked.
Given a e-commerce site, for the query "ipad" i obtain:
ipad case for ipad 2
ipad case
ipad connection kit
ipad 32gb wifi
This is a problem, since we want to rank first the main products (or products by itself) and tf/idf ranks first the accessories due to descriptions like "ipad case compatible with ipad, ipad2, ipad3, ipad retina, ipad mini, etc".
Furthermore, using the categories we have no way of determining whether is an accessory or a product.
I wonder if using automatic classification would help. Another solution that improves this ranking (like Named Entity Recognition) would be appreciated.
Could you provide tagged data?
If you have >50k items a Naive Bayes with a bigram language model trained on the product name will almost catch all accessories with 99% accuracy. I guess you can train such a naive bayes with Mahout, however product names have a pretty limited bigram amount so this can be trained even on a smartphone easily and fast nowadays.
This is a typical mechanical turk task, shouldn't be that expensive to tag a few items. However if you insist on some semi-supervised algorithm, I found Iterative similarity aggregation pretty useful.
The main idea is that you give a few tokens like "case"/"power adapter" and it iteratively finds new tokens that are indicators of spam because they appear in the same context.
Here is the paper, but I have written a blogpost about this as well which sums up the intention in plain language. This paper also mentions the same "let the user find the right item" paradigm that Sean has proposed, so both can be used in conjunction.
Oh and if you need some advice of machine learning with Lucene&SOLR I can recommend you the talk of my friend Tommaso Teofili at ApacheCon Europe this year. You can find the slides on slideshare. There is also a youtube video of the talk out there, just search for it ;)
TF/IDF is just going to rank based on the words in the query vs words in the title as you have found. That sounds like it is not the right definition of "good result" and that you want to favor products over accessories.
Of course you can simply attach heuristics to patch the problem. For example, consider the title as a set of words, not multiset, so the appearance of "iPad" several times makes no difference. Or just boost the score of items that you know are products. This isn't learning per se, but are simple, directly reflect your business knowledge, and probably have some positive effect.
If you want to learn here, you probably need to use the one best source of knowledge about what the best results are: your users. You know what they click in response to each query. You can learn a term-item model that associates search terms to items clicked. You can view that as many types of problem -- actually a latent-factor recommender model could work well there.
Have a look at Ted's slides on how to use a recommender as a "search engine": http://www.slideshare.net/tdunning/search-as-recommendation

What's the best way to store unstructured text file for data mining

I have millions of text news on my machine. I want to do some text mining on it.
I want first to store thest text news in a more structured way. what's the best way to do it ? so It will become more convenient to do data mining later on.
Currently I just store these news file in database indexed by the news headlines and the file path.
Any suggestion will be really appreciated. Thanks!
That depends greatly on what you want to achive with the more structured data.
If the data size is not heavy, you could use "in text" search on your database and you are aldready done.
A category or "tag" like here on stackoverflow would help greatly to categorize and group your content, but I guess it is very hard to extract that from your pure text base now.
Also a simple timestamp (you could get from the file itself, but be wary some systems alter that date when files get copied...) could help too.
For content extraction, have a look at http://www.opencalais.com/ , it offers an api for "text" analysis you might find interesting.
What do you mean by "do some text mining"? Are you just looking to store the text? Or, are you looking for a solution?
Many databases offer the capability to store text and do fast retrievals on them.
However, text mining typically covers a broader range of themes. Here are some examples:
Finding documents with similar themes.
Exposing sentiment in the documents.
Answering questions posed in natural language.
Summarizing documents.
Filling in data structures with information from documents.
Using information from documents for predictive modeling purposes.
Assigning codes to documents.
For such analyses, you would normally use text mining tools (you can look for these on kdnuggets.com, for instance). The tool then affects how the text is stored.
The last chapter of "Data Mining Techniques for Marketing, Sales, and Customer Support" is about text mining and has a very good case study on text mining applied to customer service records.
[In response to comment]
Is this an academic project or "real world"? Is the text monolingual? If so, is it English? You definitely need to do some research. Text analysis/mining has been an area of rather intense study since, at least, the time when Alan Turing proposed the Turing test in the 1930s.
As an example, I can readily think of four very different options for storing text for analysis. The first is "as is", which is most useful if you have lots of processors and memory. The second would be "grammatically", with text tagged with grammar and meanings, which is most effective if you have a team with lots of PhDs. Third is as an inverted index, which is the basic form for searching and some proximity matching. The fourth is by projecting onto an orthogonal space, using singular value decomposition (most useful if you want to use the text as input to other statistical techniques).

Feature selection and unsupervised learning for multilingual data + machine learning algorithm selection

Questions
I want to classify/categorize/cluster/group together a set of several thousand websites. There's data that we can train on, so we can do supervised learning, but it's not data that we've gathered and we're not adamant about using it -- so we're also considering unsupervised learning.
What features can I use in a machine learning algorithm to deal with multilingual data? Note that some of these languages might not have been dealt with in the Natural Language Processing field.
If I were to use an unsupervised learning algorithm, should I just partition the data by language and deal with each language differently? Different languages might have different relevant categories (or not, depending on your psycholinguistic theoretical tendencies), which might affect the decision to partition.
I was thinking of using decision trees, or maybe Support Vector Machines (SVMs) to allow for more features (from my understanding of them). This post suggests random forests instead of SVMs. Any thoughts?
Pragmatical approaches are welcome! (Theoretical ones, too, but those might be saved for later fun.)
Some context
We are trying to classify a corpus of many thousands of websites in 3 to 5 languages (maybe up to 10, but we're not sure).
We have training data in the form of hundreds of websites already classified. However, we may choose to use that data set or not -- if other categories make more sense, we're open to not using the training data that we have, since it is not something we gathered in the first place. We are on the final stages of scraping data/text from websites.
Now we must decide on the issues above. I have done some work with the Brown Corpus and the Brill tagger, but this will not work because of the multiple-languages issue.
We intend to use the Orange machine learning package.
According to the context you have provided, this is a supervised learning problem.
Therefore, you are doing classification, not clustering. If I misunderstood, please update your question to say so.
I would start with the simplest features, namely tokenize the unicode text of the pages, and use a dictionary to translate every new token to a number, and simply consider the existence of a token as a feature.
Next, I would use the simplest algorithm I can - I tend to go with Naive Bayes, but if you have an easy way to run SVM this is also nice.
Compare your results with some baseline - say assigning the most frequent class to all the pages.
Is the simplest approach good enough? If not, start iterating over algorithms and features.
If you go the supervised route, then the fact that the web pages are in multiple languages shouldn't make a difference. If you go with, say lexical features (bag-o'-words style) then each language will end up yielding disjoint sets of features, but that's okay. All of the standard algorithms will likely give comparable results, so just pick one and go with it. I agree with Yuval that Naive Bayes is a good place to start, and only if that doesn't meet your needs that try something like SVMs or random forests.
If you go the unsupervised route, though, the fact that the texts aren't all in the same language might be a big problem. Any reasonable clustering algorithm will first group the texts by language, and then within each language cluster by something like topic (if you're using content words as features). Whether that's a bug or a feature will depend entirely on why you want to classify these texts. If the point is to group documents by topic, irrespective of language, then it's no good. But if you're okay with having different categories for each language, then yeah, you've just got as many separate classification problems as you have languages.
If you do want a unified set of classes, then you'll need some way to link similar documents across languages. Are there any documents in more that one language? If so, you could use them as a kind of statistical Rosetta Stone, to link words in different languages. Then, using something like Latent Semantic Analysis, you could extend that to second-order relations: words in different languages that don't ever occur in the same document, but which tend to co-occur with words which do. Or maybe you could use something like anchor text or properties of the URLs to assign a rough classification to documents in a language-independent manner and use that as a way to get started.
But, honestly, it seems strange to go into a classification problem without a clear idea of what the classes are (or at least what would count as a good classification). Coming up with the classes is the hard part, and it's the part that'll determine whether the project is a success or failure. The actual algorithmic part is fairly rote.
Main answer is: try different approaches. Without actual testing it's very hard to predict what method will give best results. So, I'll just suggest some methods that I would try first and describe their pros and cons.
First of all, I would recommend supervised learning. Even if the data classification is not very accurate, it may still give better results than unsupervised clustering. One of the reasons for it is a number of random factors that are used during clustering. For example, k-means algorithm relies on randomly selected points when starting the process, which can lead to a very different results for different program runnings (though x-means modifications seems to normalize this behavior). Clustering will give good results only if underlying elements produce well separated areas in the feature space.
One of approaches to treating multilingual data is to use multilingual resources as support points. For example, you can index some Wikipedia's articles and create "bridges" between same topics in different languages. Alternatively, you can create multilingual association dictionary like this paper describes.
As for methods, the first thing that comes to mind is instance-based semantic methods like LSI. It uses vector space model to calculate distance between words and/or documents. In contrast to other methods it can efficiently treat synonymy and polysemy. Disadvantage of this method is a computational inefficiency and leak of implementations. One of the phases of LSI makes use of a very big cooccurrence matrix, which for large corpus of documents will require distributed computing and other special treatment. There's modification of LSA called Random Indexing which do not construct full coocurrence matrix, but you'll hardly find appropriate implementation for it. Some time ago I created library in Clojure for this method, but it is pre-alpha now, so I can't recommend using it. Nevertheless, if you decide to give it a try, you can find project 'Clinch' of a user 'faithlessfriend' on github (I'll not post direct link to avoid unnecessary advertisement).
Beyond special semantic methods the rule "simplicity first" must be used. From this point, Naive Bayes is a right point to start from. The only note here is that multinomial version of Naive Bayes is preferable: my experience tells that count of words really does matter.
SVM is a technique for classifying linearly separable data, and text data is almost always not linearly separable (at least several common words appear in any pair of documents). It doesn't mean, that SVM cannot be used for text classification - you still should try it, but results may be much lower than for other machine learning tasks.
I haven't enough experience with decision trees, but using it for efficient text classification seems strange to me. I have seen some examples where they gave excellent results, but when I tried to use C4.5 algorithm for this task, the results were terrible. I believe you should get some software where decision trees are implemented and test them by yourself. It is always better to know then to suggest.
There's much more to say on every topic, so feel free to ask more questions on specific topic.

Tips on how to improve full text search for search engine

I'm developing: http://www.buscatiendas.com.mx
I've seen people entering text for queries with lots of typos.
What kind of search could i implement so similar words are found?
Like google does more or less would be neat.
I'm using SQL Server Full Text search.
Why don't you have google/bing index it for you and just use that using the site: feature provided by them?
If that is not an option, you might have to have one of your own 'spell checkers' (either implement yourself or just use an existing one), which is trained on the data you have. Note spell checking is not deterministic (for eg: latel, is it label? later?). You can only make a 'best' guess based on the actual data you have in your site.
There are probabilistic models where you can 'train' your spell guesser/checker to come up with the a 'best' guess.
The following page seems pretty useful. It has a description on how to write one yourself, and also has good links (including a survey paper) and links to implementations in different languages:
http://norvig.com/spell-correct.html.
There are two ways to solve this:
Buy a 3rd party product, like a google search applicance, or one of
Microsoft search servers.
Log all queries, and have someone review these, making a table which
links the bad queries to what they
should be. (It's possible you could
buy a component library which does
this, much like a
spelling checker.)
if you want to roll out your own, first u need to filter out noise words before u even start searching because this may just impose load on your database unnecessarily. should "a good book" be the same as searching for "the good book" or "his good book" or "good and bad reviews on a book"? so obviously, "a", "the", "an", "and", etc. do not at at all qualify as "useful" search keywords. once u got the "noise" filtered out, then u start the real searching. again, u should consider database performance. is it wise to search a dynamic database or a pre-precessed database? figure out a way to filter out the noise words in the search data too.

Template Matching for relational database

I am trying to do the following:
we are trying to design a fraud detection system for stock market.
I know the Specification for the frauds (they are like templates).
so I want to know if I can design a template, and find all records that match this template.
Notice:
I can't use the traditional queries cause the templates are complex
for example one of my Fraud is circular trading,it's like this :
A bought from B, and B bought from C, And C bought from A (it's a cycle)
and this cycle can include 4 or 5 persons.
is there any good suggestion for this situation.
I don't see why you can't use "traditional queries" as you've stated. SQL can be used to write extraordinarily complex queries. For that matter I'm not sure that this is a hugely challenging question.
Firstly, I'd look at the behavior you have described as vary transactional, therefore I treat the transactions as a model. I'd likely have a transactions table with some columns like buyer, seller, amount, etc...
You could alternatively have the shares as its own table and store say the previous 100 owners of that share in the same table using STI (Single Table Inheritance) buy putting all the primary keys of the owners into an "owners" column in your shares table like 234/823/12334/1234/... that way you can do complex queries and see if that share was owned by the same person or look for patterns in the string really easily and quickly.
-update-
I wouldn't suggest making up a "small language" I don't see why you'd want to do something like that when you have huge selection of wonderful languages and databases to choose from, all of which have well refined and tested methods to solve exactly what you are doing.
My best advice is pop open your IDE (thumbs up for TextMate) and pick your favorite language (Ruby in my case). Find some sample data and create your database and start writing some code! You can't go wrong trying to experiment like this, it'll will totally expose better ways to go about it than we can dream up here on Stackoverflow.
Definitely Data Mining. But as you point out, you've already got the models (your templates). Look up fraud DETECTION rather than prevention for better search results?
I know a some banks use SPSS PASW Modeler for fraud detection. This is very intuitive and you can see what you are doing as you play around with the data. So you can implement your templates. I agree with Joseph, you need to get playing, making some new data structures.
Maybe a timeseries model?
Theoretically you could develop a "Small Language" first, something with a simple syntax (that makes expressing the domain - in your case fraud patterns - easy) and from it generate one or more SQL queries.
As most solutions, this could be thought of as a slider: at one extreme there is the "full Fraud Detection Language" at the other, you could just build stored procedures for the most common cases, and write new stored procedures which use the more "basic" blocks you wrote before to implement the various patterns.
What you are trying to do falls under the Data Mining umbrella, so you could also try to learn more about it: maybe you can find a Data Mining package for your specific DB (you didn't specify) and see if it helps you finding common patterns in your data.

Resources