I am trying named entity recognition for the first time. I'm looking for features that will pick out English names. I am using the methods outlined in the coursera nlp course (week three) and the nltk book. In other words: I am defining features, identifying features of words and then running those words/features through a classifier that I train on labeled data.
What features are used to pick out English names?
I can imagine that you'd look for two capital words in a row, or a capital word and then an initial and then a capital word. (ex. John Smith or James P. Smith).
But what other features are used for NER?
Some common features:
Word lists for common names (John, Adam, etc)
casing
contains symbol or numeric characters (names generally don't)
person prefixes (Mr., Mrs., etc...)
person postfixes (Jr., Sr., etc...)
single letter abbreviation (ie, (J.) Smith).
analysis of surrounding words (you may find some words have a high probability of appearing near names).
Named Entities previously recognized (often it is easy to identify NE in some parts of the corpus based on context, but very hard in other parts. If previously identified, this is an excellent hint towards NER)
Depending what language you are working with there may be more language specific features as well. Frankly you can turn up a wealth of information with a simple Google query, I'm really not sure why you haven't turned there. Some starting points however:
Google
A survey of named entity recognition and classification
Named entity recognition without gazetteers
I had done something similar back in school using machine learning. I suppose that you will use a supervised algorithm and you will classify every single word independently and not words in combination. In that case I would choose some features for the word itself like the ones you mentioned (if the word begins with a capital letter, if the word is an abbreviation) but I would add some more features like if the previous or the next words also start from a capital letter, or if they are abbreviations. This way you can add some context and overcome the problems related to your basic independence assumption.
If you want have a look here. In the machine learning section you can find some more information and examples (the problem is slightly different but the method should be similar).
Whatever features you choose it is important that you use some measure to evaluate their relevance and possibly reduce them to the useful ones to avoid over-fitting. One of the measures you can use to evaluate them is the gain ratio but there are many more. Here you can find some basic information about feature extraction.
Hope it helps!
Related
I have a database of several thousands of utterances. Each record (utterance) is a text representing a problem description, which a user has submitted to a service desk. Sometimes also the service desk agent's response is included. The language is highly technical, and it contains three types of tokens:
words and phrases in Language 1 (e.g. English)
words and phrases in Language 2 (e.g. French, Norwegian, or Italian)
machine-generated output (e.g. listing of files using unix command ls -la)
These languages are densely mixed. I often see that in one conversation, a sentence in Language 1 is followed by Language 2. So it is impossible to divide the data into two separate sets, corresponding to utterances in two languages.
The task is to find similarities between the records (problem descriptions). The purpose of this exercise is to understand whether some bugs submitted by users are similar to each other.
Q: What is the standard way to proceed in such a situation?
In particular, the problem lies in the fact that the words come from two different corpora (corpuses), while in addition, some technical words (like filenames, OS paths, or application names) will not be found in any.
I don't think there's a "standard way" - just things you could try.
You could look into word-embeddings that are aligned between langauges – so that similar words across multiple languages have similar vectors. Then ways of building a summary vector for a text based on word-vectors (like a simple average of all a text's words' vectors), or pairwise comparisons based on word vectors (like "Word Mover's Distance"), may still work with mixed-language texts (even mixes of languages within one text).
That a single text, presumably about a a single (or closely related) set of issues, has mixed language may be a blessing rather than a curse: some classifiers/embeddings you train from such texts might then be able to learn the cross-language correlations of words with shared topics. But also, you could consider enhancing your texts with extra synthetic auto-translated text, for any monolingual ranges, to ensure downstream embeddings/comparisons get closer to your ideal of language-obliviousness.
Thank you for the suggestions. After several experiments I developed a method which is simple and works pretty well. Rather than using existing corpora, I created my own corpus based on all the utterances available in my multilingual database. Without translating them. The database has 130,000 utterances, including 3,5 million of words (in three languages: English, French and Norwegian) and 150,000 unique words. The phrase similarity based on the meaning space constructed this way works surprisingly well. I have tested this method on production and the results are good. I also see a lot of space for improvement, and will continue to polish it. I also wrote this article An approach to categorize multi-lingual phrases, describing all the steps in more detail. Critics or improvements welcome.
I'm working on a project where I am trying to build a named entity recognizer from texts. So basically I want to build and experiment the NER in 3 different ways.
First, I want to build it using only segmented sentences-> tokenized words. To clarify, I want to input only split/tokenized words into the system. Once again, the NER system is rule-based. Hence, it can only use rules to conclude which is a named entity. In the first NER, it will not have any chunk information or part of speech label. Just the tokenized words. Here, the efficiency is not the concern. Rather the concern lies in comparing the 3 different NERs, how they perform. (The one I am asking about is the 1st one).
I thought of it for a while and could not figure out any rules or any idea of coming up with a solution to this problem. One naive approach would be to conclude all words beginning with an uppercase and that does not follow a period to be a named entity.
Am I missing anything? Any heads up or guidelines would help.
Typically NER relies on preprocessing such as part-of-speech tagging (named entities are typically nouns), so not having this basic information makes the task more difficult and therefore more prone to error. There will be certain patterns that you could look for, such as the one you suggest (although what do you do with sentence-initial named entities?). You could add certain regular expression patterns with prepositions, e.g. (Title_case_token)+ of (the)? (Title_case_token)+ would match "Leader of the Free World", "Prime Minister of the United Kindom", and "Alexander the Great". You might also want to consider patterns to match acronyms such as "S.N.C.F.", "IBM", "UN", etc. A first step is probably to look for some lexical resources (i.e. word lists) like country names, first names, etc., and build from there.
You could use spaCy (Python) or TokensRegex (Java) to do token-based matching (and not use the linguistic features they add to the tokens).
I'm embarking on a project for a non-profit organization to help process and classify 1000's of reports annually from their field workers / contractors the world over. I'm relatively new to NLP and as such wanted to seek the group's guidance on the approach to solve our problem.
I'll highlight the current process, and our challenges and would love your help on the best way to solve our problem.
Current process: Field officers submit reports from locally run projects in the form of best practices. These reports are then processed by a full-time team of curators who (i) ensure they adhere to a best-practice template and (ii) edit the documents to improve language/style/grammar.
Challenge: As the number of field workers increased the volume of reports being generated has grown and our editors are now becoming the bottle-neck.
Solution: We would like to automate the 1st step of our process i.e., checking the document for compliance to the organizational best practice template
Basically, we need to ensure every report has 3 components namely:
1. States its purpose: What topic / problem does this best practice address?
2. Identifies Audience: Who is this for?
3. Highlights Relevance: What can the reader do after reading it?
Here's an example of a good report submission.
"This document introduces techniques for successfully applying best practices across developing countries. This study is intended to help low-income farmers identify a set of best practices for pricing agricultural products in places where there is no price transparency. By implementing these processes, farmers will be able to get better prices for their produce and raise their household incomes."
As of now, our approach has been to use RegEx and check for keywords. i.e., to check for compliance we use the following logic:
1 To check "states purpose" = we do a regex to match 'purpose', 'intent'
2 To check "identifies audience" = we do a regex to match with 'identifies', 'is for'
3 To check "highlights relevance" = we do a regex to match with 'able to', 'allows', 'enables'
The current approach of RegEx seems very primitive and limited so I wanted to ask the community if there is a better way to solving this problem using something like NLTK, CoreNLP.
Thanks in advance.
Interesting problem, i believe its a thorough research problem! In natural language processing, there are few techniques that learn and extract template from text and then can use them as gold annotation to identify whether a document follows the template structure. Researchers used this kind of system for automatic question answering (extract templates from question and then answer them). But in your case its more difficult as you need to learn the structure from a report. In the light of Natural Language Processing, this is more hard to address your problem (no simple NLP task matches with your problem definition) and you may not need any fancy model (complex) to resolve your problem.
You can start by simple document matching and computing a similarity score. If you have large collection of positive examples (well formatted and specified reports), you can construct a dictionary based on tf-idf weights. Then you can check the presence of the dictionary tokens. You can also think of this problem as a binary classification problem. There are good machine learning classifiers such as svm, logistic regression which works good for text data. You can use python and scikit-learn to build programs quickly and they are pretty easy to use. For text pre-processing, you can use NLTK.
Since the reports will be generated by field workers and there are few questions that will be answered by the reports (you mentioned about 3 specific components), i guess simple keyword matching techniques will be a good start for your research. You can gradually move to different directions based on your observations.
This seems like a perfect scenario to apply some machine learning to your process.
First of all, the data annotation problem is covered. This is usually the most annoying problem. Thankfully, you can rely on the curators. The curators can mark the specific sentences that specify: audience, relevance, purpose.
Train some models to identify these types of clauses. If all the classifiers fire for a certain document, it means that the document is properly formatted.
If errors are encountered, make sure to retrain the models with the specific examples.
If you don't provide yourself hints about the format of the document this is an open problem.
What you can do thought, is ask people writing report to conform to some format for the document like having 3 parts each of which have a pre-defined title like so
1. Purpose
Explains the purpose of the document in several paragraph.
2. Topic / Problem
This address the foobar problem also known as lorem ipsum feeling text.
3. Take away
What can the reader do after reading it?
You parse this document from .doc format for instance and extract the three parts. Then you can go through spell checking, grammar and text complexity algorithm. And finally you can extract for instance Named Entities (cf. Named Entity Recognition) and low TF-IDF words.
I've been trying to do something very similar with clinical trials, where most of the data is again written in natural language.
If you do not care about past data, and have control over what the field officers write, maybe you can have them provide these 3 extra fields in their reports, and you would be done.
Otherwise; CoreNLP and OpenNLP, the libraries that I'm most familiar with, have some tools that can help you with part of the task. For example; if your Regex pattern matches a word that starts with the prefix "inten", the actual word could be "intention", "intended", "intent", "intentionally" etc., and you wouldn't necessarily know if the word is a verb, a noun, an adjective or an adverb. POS taggers and the parsers in these libraries would be able to tell you the type (POS) of the word and maybe you only care about the verbs that start with "inten", or more strictly, the verbs spoken by the 3rd person singular.
CoreNLP has another tool called OpenIE, which attempts to extract relations in a sentence. For example, given the following sentence
Born in a small town, she took the midnight train going anywhere
CoreNLP can extract the triple
she, took, midnight train
Combined with the POS tagger for example; you would also know that "she" is a personal pronoun and "took" is a past tense verb.
These libraries can accomplish many other tasks such as tokenization, sentence splitting, and named entity recognition and it would be up to you to combine all of these tools with your domain knowledge and creativity to come up with a solution that works for your case.
I have a collection of bills and Invoices, so there is no context in the text (i mean they don't tell a story).
I want to extract people names from those bills.
I tried OpenNLP but the quality of trained model is not good because i don't have context.
so the first question is: can I train model contains only people names without context? and if that possible can you give me good article for how i build that new model (most of the article that i read didn't explain the steps that i should made to build new model).
I have database name with more than 100,000 person name (first name, last name), so if the NER systems don't work in my case (because there is no context), what is the best way to search for those candidates (I mean searching for every first name with all other last names?)
thanks.
Regarding "context", I guess you mean that you don't have entire sentences, i.e. no previous / next tokens, and in this case you face quite a non-standard NER. I am not aware of available software or training data for this particular problem, if you found none you'll have to build your own corpus for training and/or evaluation purposes.
Your database of names will probably greatly help, depending indeed on what proportion of bill names are actually present in the database. You'll also probably have to rely on character-level morphology of names, as patterns (see for instance patterns in [1]). Once you have a training set with features (presence in database, morphology, other information of bill) and solutions (actual names of annotated bills), using standard machine-learning as SVM will be quite straightforward (if you are not familiar with this, just ask).
Some other suggestions:
You may most probably also use other bill's information: company name, positions, tax mentions, etc.
You may also proceed in a a selective manner - if all bills should mention (exactly?) one person name, you may exclude all other texts (e.g. amounts, tax names, positions etc.) or assume in a dedicated model that among all text in a bill, only one should be guessed as a name.
[1] Ranking algorithms for named-entity extraction: Boosting and the voted perceptron (Michael Collins, 2002)
I'd start with some regular expressions, then possibly augment that with a dictionary-based approach (i.e., big list of names).
No matter what you do, it won't be perfect, so be sure to keep that in mind.
If I have a given text (both long or short), with which methods do you usually detect which language it is written in?
It is clear that:
You need a training corpus to train the models you use (e.g. neural networks, if used)
Easiest thing coming to my mind is:
Check characters used in the text (e.g. hiragana are only used in Japanese, Umlauts probably only in European languages, ç in French, Turkish, …)
Increase the check to two or three letter pairs to find specific combinations of a language
Lookup a dictionary to check which words occur in which language (probably only without stemming, as stemming depends on the language)
But I guess there are better ways to go. I am not searching for existing projects (those questions have already been answered), but for methods like Hidden-Markov-Models, Neural Networks, … whatever may be used for this task.
In product I'm working on we use dictionary-based approach.
First relative probabilities for all words in training corpus are calculated and this is stored as a model.
Then input text is processed word by word to see if particular model gives best match (much better then the other models).
In some cases all models provide quite bad match.
Few interesting points:
As we are working with social media both normalized and non-normalized matches are attempted (in this context normalization is removal of diacritics from symbols). Non-normalized matches have a higher weight
This method works rather bad on very short phrases (1-2 words) in particular when these words are there in few languages, which is the case of few European languages
Also for a better detection we are considering added per-character model as you have described (certain languages have certain unique characters)
Btw, we use ICU library to split words. Works rather good for European and Eastern languages (currently we support Chinese)
Check the Cavnar and Trenkle algorithm.