Okay, I've seen lots of posts about fuzzy string matching, Levenstein distance, longest common substring, and so on. None of them seem to fit exactly for what I want to do. I'm pulling product results from a variety of web services, and from those I can build a big list of names for the product. These names may include a bunch of variable junk. Here are some examples, from SearchUPC:
Apple 60W magsafe adapter L-shape with extension cord
Original Apple 60W Power Adapter (L-shaped Connector) for MacBook MC461LL/A with AC Extension Wall Cord (Bulk Packaging)
Current Apple MagSafe 60W Power Adapter for MacBook MC461LL/A with AC Extension Wall Cord (Bulk Packaging)
Apple 60W MagSafe Power Adapter - Apple Mac Accessories
Apple - MagSafe 60W Power Adapter for MacBook and 13\" MacBook Pro
MagSafe - power adapter - 60 Watt
etc. What I'd like to do is pull the common product name (which to my heuristic human eye is obviously Apple 60W MagSafe Power Adapter), but none of the aforementioned methods seem likely to work. My main problem is that I don't know what to search the list of strings for... At first, I was thinking of trying longest common substring, but it seems like that will fail as a bunch of the strings have things out of order, which might yield a product name of power adapter, which is not terribly useful to the user.
Note: the vast majority of the records returned from the SearchUPC API (mostly omitted here) do include the literal string "Apply 60W MagSafe Power Adapter".
I'm implementing this in Objective-C, for iOS, but I'm really interested in the algorithm more than the implementation, so any language is acceptable.
If you want to compare strings but need something more robust than longest common substring with respect to changed order of substrings, you might look into a technique called string tiling. Simplified, the principle is as follows:
Find the largest common substring (larger than a minimum length) in both strings
Remove that substring from both strings
Repeat until there are no more substrings larger than your minimal length
In practice, the ratio between the remaining (unmatched) stringparts to the initial lengths is a excellent indicator on how well the strings match. And the technique is extremely robust against reordering of substrings. You can find the scientific paper by M. Wise describing this technique here. I have implemented the algorithm myself in the past (it's not hard), but apparantly free implementations are readily available (for example here). While I have used the algorithm in a variety of fuzzy matching scenarios, I can't vouch for the implementation which I have never used myself.
String tiling does not solve the problem of finding the largest common productname in itself, but it seems to me that by slightly modifying the algorithm, you can do so. You're probably less interested in calculating the matchpercentage than in the keeping the similar parts?
As the previous poster remarked, such fuzzy matching almost always needs some kind of manual validation, as both false positives and false negatives are unavoidable.
Sounds like your approach needs to be two fold. One should be matching records with one another. The other part is pulling the "canonical name" out of those matching records. In your case it is a product, but same concept. I'm not sure how you will go about associating groups of matching records with the standardized product name, but I would suggest trying to pull the important information out of the records and trying to match that to some resources on the Internet. For instance, for your example, maybe you compare your data to an apple product list. Or you could try and have a robot that crawls google and pulls the top result to try and associate. Bottom line, at the end of the day, if your text is really dirty, you will need to include some human intervention. What I mean is that you may set a threshold for match, no match, and needs review. Good luck.
Related
I'm new to natural language process so I apologize if my question is unclear. I have read a book or two on the subject and done general research of various libraries to figure out how i should be doing this, but I'm not confident yet that know what to do.
I'm playing with an idea for an application and part of it is trying to find product mentions in unstructured text (e.g. tweets, facebook posts, emails, websites, etc.) in real-time. I wont go into what the products are but it can be assumed that they are known (stored in a file or database). Some examples:
"starting tomorrow, we have 5 boxes of #hersheys snickers available for $5 each - limit 1 pp" (snickers is the product from the hershey company [mentioned as "#hersheys"])
"Big news: 12-oz. bottles of Coke and Pepsi on sale starting Fri." (coca-cola is the product [aliased as "coke"] from coca-cola company and Pepsi is the product from the PepsiCo company)
"#OMG, i just bought my dream car. a mustang!!!!" (mustang is the product from Ford)
So basically, given a piece of text, query the text to see if it mentions a product and receive some indication (boolean or confidence number) that it does mention the product.
Some concerns I have are:
Missing products because of misspellings. I thought maybe i could use a string similarity check to catch these.
Product names that are also English words or things would get caught. Like mustang the horse versus mustang the car
Needing to keep a list of alternative names for products (e.g. "coke" for "coco-cola", etc.)
I don't really know where to start with this but any help would be appreciated. I've already looked at NLTK and SciKit and didn't really gleam how to do this from there. If you know of examples or papers that explain this, links would be helpful. I'm not specific to any language at this point. Java preferably but Python and Scala are acceptable.
The answer that you chose is not really answering your question.
The best approach you can take is using Named Entity Recognizer(NER) and POS tagger (grab NNP/NNPS; Proper nouns). The database there might be missing some new brands like Lyft (Uber's rival) but without developing your own prop database, Stanford tagger will solve half of your immediate needs.
If you have time, I would build the dictionary that has every brands name and simply extract it from tweet strings.
http://www.namedevelopment.com/brand-names.html
If you know how to crawl, it's not a hard problem to solve.
It looks like your goal is to classify linguistic forms in a given text as references to semantic entities (which can be referred to by many different linguistic forms). You describe a number of subtasks which should be done in order to get good results, but they nevertheless are still independent tasks.
Misspellings
In order to deal with potential misspellings of words, you need to associate these possible misspellings to their canonical (i.e. correct) form.
Phonetic similarity: Many reasons for "misspellings" is opacity in the relationship between the word's phonetic form (i.e. how it sounds) and its orthographic form (i.e. how it's spelled). Therefore, a good way to address this is to index terms phonetically so that e.g. innovashun is associated with innovation.
Form similarity: Additionally, you could do a string similarity check, but you may introduce a lot of noise into your results which you would have to address because many distinct words are in fact very similar (e.g. chic vs. chick). You could make this a bit smarter by first morphologically analyzing the word and then using a tree kernel instead.
Hand-made mappings: You can also simply make a list of common misspelling → canonical_form mappings. This would work well for "exceptions" not handled by the above methods.
Word-sense disambiguation
Mustang the car and Mustang the horse are the same form but refer to entirely different entities (or rather classes of entities, if you want to be pedantic). In fact, we ourselves as humans can't tell which one is meant unless we also know the word's context. One widely-used way of modelling this context is distributional lexical semantics: Defining a word's semantic similarity to another as the similarity of their lexical contexts, i.e. the words preceding and succeeding them in text.
Linguistic aliases (synonyms)
As stated above, any given semantic entity can be referred to in a number of different ways: bathroom, washroom, restroom, toilet, water closet, WC, loo, little boys'/girls' room, throne room etc. For simple meanings referring to generic entities like this, they can often be considered to be variant spellings in the same way that "common misspellings" are and can be mapped to a "canonical" form with a list. For ambiguous references such as throne room, other metrics (such as lexical-distributional methods) can also be included in order to disambiguate the meaning, so that you don't relate e.g. I'm in the throne room just now! to The throne room of the Buckingham Palace is beautiful.
Conclusion
You have a lot of work to do in order to get where you want to go, but it's all interesting stuff and there are already good libraries available for doing most of these tasks.
What are the recommended methods for extracting locations from free text?
What I can think of is to use regex rules like "words ... in location". But are there better approaches than this?
Also I can think of having a lookup hash table table with names for countries and cities and then compare every extracted token from the text to that of the hash table.
Does anybody know of better approaches?
Edit: I'm trying to extract locations from tweets text. So the issue of high number of tweets might also affect my choice for a method.
All rule-based approaches will fail (if your text is really "free"). That includes regex, context-free grammars, any kind of lookup... Believe me, I've been there before :-)
This problem is called Named Entity Recognition. Location is one of the 3 most studied classes (with Person and Organization). Stanford NLP has an open source Java implementation that is extremely powerful: http://nlp.stanford.edu/software/CRF-NER.shtml
You can easily find implementations in other programming languages.
Put all of your valid locations into a sorted list. If you are planning on comparing case-insensitive, make sure the case of your list already is normalized.
Then all you have to do is loop over individual "words" in your input text and at the start of each new word, start a new binary search in your location list. As soon as you find a no-match, you can skip the entire word and proceed with the next.
Possible problem: multi-word locations such as "New York", "3rd Street", "People's Republic of China". Perhaps all it takes, though, is to save the position of the first new word, if you find your bsearch leads you to a (possible!) multi-word result. Then, if the full comparison fails -- possibly several words later -- all you have to do is revert to this 'next' word, in relation to the previous one where you started.
As to what a "word" is: while you are preparing your location list, make a list of all characters that may appear inside locations. Only phrases that contain characters from this list can be considered a valid 'word'.
How fast are the tweets coming in? As in is it the full twitter fire hose or some filtering queries?
A bit more sophisticated approach, that is similar to what you described is using an NLP tool that is integrated to a gazetteer.
Very few NLP tools will keep up to twitter rates, and very few do very well with twitter because of all of the leet speak. The NLP can be tuned for precision or recall depending on your needs, to limit down performing lockups in the gazetteer.
I recommend looking at Rosoka(also Rosoka Cloud through Amazon AWS) and GeoGravy
Help by editing my question title and tags is greatly appreciated!
Sometimes one participant in my corpus of "conversations" will refer to another participant using a nickname, usually an abbreviation or misspelling, but hereafter I'll just say "nicknames". Let's say I'm willing to manually tell my software whether or not I think various possible nicknames are in fact nicknames, but I want software to come up with a list of possible matches between the handle's that identify people, and the potential nicknames. How would I go about doing that?
Background on me and then my corpus: I have no experience doing natural language processing but I'm a competent data analyst with R. My data is produced by 70 teams, each forecasting the likelihood of 100 distinct events occurring some time in the future. The result that I have 70 x 100 = 7000 text files, containing the stream of forecasts participants make and the comments they include with their forecasts. I'll paste a very short snip of one of these text files below, this one had to do with whether the Malian government would enter talks with the MNLA:
02/12/2013 20:10: past_returns answered Yes: (50%)
I hadn't done a lot of research when I put in my previous
placeholder... I'm bumping up a lot due to DougL's forecast
02/12/2013 19:31: DougL answered Yes: (60%)
Weak President Traore wants talks if MNLA drops territorial claims.
Mali's military may not want talks. France wants talks. MNLA sugggests
it just needs autonomy. But in 7 weeks?
02/12/2013 10:59: past_returns answered No: (75%)
placeholder forecast...
http://www.irinnews.org/Report/97456/What-s-the-way-forward-for-Mali
My initial thoughts: Obviously I can start by providing the names I'm looking to match things up with... in the above example they would be past_returns and DougL (though there is no use of nicknames in the above). I wouldn't think it'd be that hard to get a computer to guess at minor misspellings (though I wouldn't personally know where to start). I can imagine that other tricks could be used, like assuming that a string is more likely to be a nickname if it is used much much more by one team, than by other teams. A nickname is more likely to refer to someone who spoke recently than someone who spoke long ago, or not at all on regarding this question. And they should be used in sentences in a manner similar to the way the full name/screenname is typically used in the corpus. But I'm interested to hear about simple approaches, as well as ones that try to consider more sophisticated techniques.
This could get about as complicated as you want to make it. From the semi-linguistic side of things, research topics would include Levenshtein Distance (for detecting minor misspellings of known names/nicknames) and Named Entity Recognition (for the task of detecting names/nicknames in the first place). Actually, NER's worth reading about, but existing systems might not help you much in your domain of forum handles and nicknames.
The first rough idea that comes to mind is that you could run a tokenized version of your corpus against an English dictionary (perhaps a dataset compiled from Wiktionary or something like WordNet) to find words that are candidates for names, then filter those through some heuristics (do they start with the same letters as known full names? Do they have a low Levenshtein distance from known names? Are they used more than once?).
You could also try some clustering or supervised ML algorithms against the non-word tokens. That might reveal some non-"word" tokens that often occur in the same threads as a given username; again, heuristics could help rule out some false positives.
Good luck; sounds like a fun problem - hope I mentioned at least one thing you hadn't already thought of.
Suppose I have a set of phrases - about 10 000 - of average length - 7-20 words in which I want to find some given phrase. The phrase I am looking for could have some errors - for example miss one or two words, have some words misplaced, or some random words - for example my database contains "As I was riding my red bike, I saw Christine", and I want it to much "As I was riding my blue bike, saw Christine", or "I was riding my bike, I saw Christine and Marion". What could be some good approach to this problem? I know about Levenhstein's distance, and I also suppose that this problem may have no easy, good solution.
A good text search engine will provide capabilities such as you describe, fsh. A typical approach would be to create a query that matches if any of the words occurs and orders the results using a weight based on number of terms occurring in proximity to each other and weighted inversely to their probability of occurring, since uncommon words will be less likely to co-occur by chance. There's a whole theory of this sort of thing called information retrieval, but maybe you know about that. Furthermore you'd like to make sure that word-level fuzziness gets accounted for by normalizing case, punctuation and the like and applying some basic linguistic transformations (stemming), and in some cases introducing a dictionary of synonyms, especially when there is domain knowledge available to condition it.
If you're interested in messing around with this stuff, try an open-source search engine, this article by Vik gives a reasonable survey from the perspective of 2009, and this one by Middleton and Baeza-Yates gives a good detailed introduction to the topic.
I'm looking for references on separating a name: "John A. Doe" in parts, first=John, middle=A., last=Doe. In Mexico we have paternal, maternal, first and second given names, and can be written in different permutations, so the problem is quite complex.
As it depends on data, we are working with matching software that calculates a score for every word so we can take decisions (it is based on a big database). The input data is not clean, it is imported from some government web pages and is human filtered so it could have junk that has to be recognized as well. Any suggestions?
[Edit]
Examples:
name:
Javier Abdul Córdoba Gándara
common permutations (or as it may appear in gvt data referring to same person):
Córdoba Gándara Javier Abdul
Javier A. Córdoba Gándara
Javier Abdul Córdoba G.
paternal=Córdoba
maternal=Gándara
first given:Javier
second given:Abdul
name: María de la Luz Sánchez Martínez
paternal:Sánchez
maternal: Martínez
first given: María de la Luz
name: Paloma Viridiana Alin Arias Medina
paternal: Arias
maternal: Medina
first given: Paloma
second given: Viridiana Alin
As I said what the meaning of each word depends on the score. One has no way of knowing that Viridiana and Alin are given names if not from the score.
We have a very strong database (80 million records or so) so we can get some use of the scoring system. I am designing some algorithm that uses that but looking for other references.
Unfortunately - and having done quite a bit of this work myself - your ideal algorithm will be very data specific, and you will need to work this out for your particular situation.
Of the total time and effort to develop this algorithm, I'd say the time will be split roughly as follows:
10% for general string manipulation
30% for the specific nature of the
data (Mexican name formats, data input quirks)
60% to cater for data quality / lack of
quality
And I believe that's quite generous towards the general string manipulation. Of course it depends whether you need quality results for all records, or only the 'clean' records etc, and if you are able to ignore the 'difficult' records it makes it a lot simpler.
Some general tips
If they are not required, remove non alphanumeric / whitespace characters
Split on spaces
Use hyphens / punctuation to identify surnames or family names
Initials (which are generally single
letters) are not surnames; i.e. they
must be first / middle
determine the level of confidence that you have programmatically identified the each name (and test this thoroughly). You may find there are subsets of data that contain similar patterns that need to be catered for individually (they may come from different sources etc)
You may need to add some natural language or machine learning to check. The problem of identifying author names (e.g. in scientific papers) is difficult as they can be reported with differing orders, degrees of abbreviation, elisions etc. If your database is dirty you will end with ambiguity whatever you do.