A few papers on the topics of word and document embeddings (word2vec, doc2vec) mention that they used the Stanford CoreNLP framework to tokenize/lemmatize/POS-tag the input words/sentences:
The corpora were lemmatized and POS-tagged with the Stanford CoreNLP (Manning et al., 2014) and each token was replaced with its lemma and POS tag
(http://www.ep.liu.se/ecp/131/039/ecp17131039.pdf)
For pre-processing, we tokenise and lowercase the words using Stanford CoreNLP
(https://arxiv.org/pdf/1607.05368.pdf)
So my questions are:
Why does the first paper apply POS-tagging? Would each token then be replaced with something like {lemma}_{POS} and the whole thing used to train the model? Or are the tags used to filter tokens?
For example, gensims WikiCorpus applies lemmatization per default and then only keeps a few types of part of speech (verbs, nouns, etc.) and gets rid of the rest. So what is the recommended way?
The quote from the second paper seems to me like they only split up words and then lowercase them. This is also what I first tried before I used WikiCorpus. In my opinion, this should give better results for document embeddings as most of POS types contribute to the meaning of a sentence. Am I right?
In the original doc2vec paper I did not find details about their pre-processing.
For your first question, the answer is "it depends on what you are trying to accomplish!"
There isn't a recommended way per say, to pre-process text. To clean a text corpus, usually the first steps are tokenization and lemmatization. Next, to remove not important terms/tokens, you can remove stop-words or even apply POS tags, to be able to remove tokens based on their grammatical category, based on the assumption that some grammatical categories (such as adjectives), do not contain valuable information for modelling a topic for example. But this purely depends on the type of analysis you are going to follow after the pre-processing step.
For you second part of the question, as explained above, tokenisation and lower case tokens, are standard parts of the pre-processing routine. So I also suspect, that regardless of the ML algorithm used later on, your results will be better if you carefully pre-process your data. I am not sure whether POS tags contribute to the meaning of a sentence though.
Hope I provided some valuable feedback to your research. If not you could provide a code sample to further discuss this issue.
Related
I have transcripts of phone calls with customers and agents. I'm trying to find promises which were made by an agent to a customer.
I already did punctuation restoration. But there are a lot of sentences that don't have any sense. I would like to remove them from the transcript. Most of them are just a set of not connected words.
I wonder what approach is the best for this task?
My ideas are:
• Use tf idf and word2vec to create vectors from all sentences. After that we can do some kind of anomaly detection e.g. look for and delete vectors that are highly deviated from most other vectors.
• Spam filters. Maybe is it possible to apply spam filters for this task?
• Crate some pattern of part of speech tags that proper sentence must include. For example, any good sentence must include noun + verb. Or we can use for example dependency tokens from spacy.
Examples
Example of a sentence that I want to keep:
There's no charge once sent that you'll get a ups tracking number.
Example of a junk sentence:
Kinder pr just have to type it in again, clock drives bethel.
Another junk sentence:
Just so you have it on and said this is regarding that.
One thing I would try is to treat this as a classification problem (junk vs non-junk). You can train a model based on a labelled set (i.e. you need to label some subset of your dataset) and then classify the rest of the corpus.
You could use a pre-trained language model like Bert and fine-tune it with you labeled set, as in here (https://colab.research.google.com/github/google-research/bert/blob/master/predicting_movie_reviews_with_bert_on_tf_hub.ipynb).
The advantage of using a language model like this is that you don't have to worry too much about linguistic (pre-)processing, meaning you don't have to get the part-of-speech or syntactic structure.
Comments regarding your ideas:
Anomaly detection with tf-idf and word2vec: It depends on the proportion of the junk sentences in your corpus. If they it's more than 15%, I would think that they might not be so anomal. Also, I am assuming your junk sentences come from noisy automatic speech-to-text transcription. I am not sure, to what extent parts of these junk sentences are correctly transcribed and what the effect of the correctly transcribed portion might have on the extent of the anomaly.
If you mean pre-existing spam filters that are trained on spam email, I would guess that the spammyness of emails is quite different from junkiness of your transcripts.
Use POS tags or syntactic structure to manually create rules for valid sentences:
This seems a bit tedious too me and also I am not sure if you will discover all junk with this. For instance, in your junk examples, the syntactic structure does not strike me as too unusal, e.g. "clock drives bethel" might be tagged as , which is quite a common tag sequence. The junkiness in this case comes from the meaning of the words.
I am using TreeTagger to get the lemmas of words in Spanish, but I have observed there are too much words which are not transformed as should be. I would like to know how this operations works, if it is done with techniques such as decision trees or machine learning algorithms or it simply contains a list of words with its corresponding lemma. Does someone know it?
Thanks!!
On basis of personal communication via email with H. Schmid, the author of TreeTagger, the answer to your question is:
The lemmatization function is based on the XTAG Project, which includes a morphological analyzer. Within the XTAG project several corpora have been analyzed. Considerung TreeTagger, especially the analysis of the Penn Treebank Corpus seems relevant, since this corpus is the training corpus for the English parameter file of TreeTagger. Considering lemmatization, the lemmata have simply been stored in a lexicon. TreeTagger finally uses this lexicon as a lookup table.
Hence, with TreeTagger you may only retreive the lemmata that are available in the lexicon.
In case you need additional funtionality regarding lemmatization beyond the options in TreeeTagger, you will need a morphological analyzer and, depending on your approach, a suitable training corpus, although this does not seem mandatoriy, since several analyzers perform quite well even when directly applied on the corpus of interest to be analyzed.
Let us say we have an article that we want to annotate. If we input the text as one really long Sentence as opposed to a Document, does Stanford do anything differently between annotating that one long Sentence as opposed to looping through every Sentence in the Document and culminating all of its results together?
EDIT: I ran a test and it seems like the two approaches return two different NER sets. I might be just doing it wrong, but it's certainly super interesting and I'm curious as to why this happens.
To confirm: you mean Stanford CoreNLP (as opposed to Apache OpenNLP), right?
The main difference in the CoreNLP Simple API between a Sentence and a Document is tokenization. A Sentence will force the entire text to be considered as a single sentence, even if it has punctuation. A Document will first tokenize the text into a list of sentences, and then annotate each sentence.
Note that for annotators like the constituency parser, very long sentences will take prohibitively long to annotate. Also, note that coreference only works on documents, not sentences.
I'm doing a keyword extraction using TD-IDF on a large number of documents. Currenly I'm splitting each sentence based on n-gram. More particularly I'm using tri-gram. However, this is not the best way to split each sentence into ints constituting keywords. For example a noun phrase like 'triple heart bypass' may not always get detected as one term.
The other alternative to chunk each sentence into its constituting elements look to be part of speech tagging and chunking in Open NLP. In this approach phrase like 'triple heart bypass' always gets extracted as a whole but the downside is in TF-IDF the frequency of extracted terms (phrases) dramatically drops.
Does anyone have any suggestion on either or these two approaches or have any other ideas to improve the quality of the keywords?
What is :
the goal of your application ?
--impacts the tokenization rules and defines the quality of your keywords
type of documents?
--chunking is not the same if you have forum data or news article data.
You can implement some boundary recognizer by yourself, or using a statistical model as in openNLP.
The typical pipeline is that you should first tokenize as simple as possible, apply stop words removal (language-dependent), and then if needed POS tagging-based filtering (but this is a costly operation).
other options : java.text.BreakIterator, com.ibm.icu.text.BreakIterator, com.ibm.icu.text.RuleBasedBreakIterator...
I want to use CRF for sentence level sentiment classiciation (positive or negative). But, I am lost on how to create a very simple feature to detect this using either CRFsuite or CRF++. Been trying for a few days, can anyone suggest how to design a simple feature which I can use as starting point to understand how to use the tools.
Thanks.
You could start providing gazetteers containing words separated by sentiment (e.g. positive adjectives, negative nouns, etc) and so using CRF to label relevant portions of the sentences. Using gazetteers you can also provide lists of other words which won't be labeled themselves, but could help identifying sentiment terms. You could also use WordNet instead of gazetteers. Your gazetteer features could be binary, i.e. gazetteer matched or not matched. Check out http://crfpp.googlecode.com for more examples and references.
I hope this helps!