# Tag Info

## New answers tagged nlp

1

I think there's a bit of confusion here: the sample you're showing is not a full "conll" format, at least not any recent one. It's simply a BIO format for NER. As far as I know conllu has been the standard "conll" format for probably at least 10 years, so if you're using some old data it's possible that it used the name "conll" ...

6

First most of the time there's no "missing text", there's an empty string (0 sentences, 0 words) and this is a valid text value. The distinction is important, because the former usually means that the information was not captured whereas the latter means that the information was intentionally left blank. For example a user not entering a review is ...

0

Transformer models such as BERT, DistilBert can be used to capture the document embeddings. Transformer models can capture the context more accurately than other models.

1

The main reason is that, normally, the tasks you pretrain and finetune the model are different, e.g. masked language modeling vs. sequence classification or tagging. This is because unlabeled data is abundant and labeled data is scarce, and pretraining on a language modeling task allows you to use a lot of data in a scarce data setup. This is, for instance, ...

0

Please be specific on what basis The problem is that the model is largely overfitting has been concluded. The validation loss decreased and did not show any evidence of overfitting. How to Fine-Tune BERT for Text Classification? demonstrated the Further Pre-training as the fine-tuning method and the diagrams of the training exhibit the similar diagram for ...

0

First need to understand what problems BERT can solve or what kind of inference/prediction it can achieve. BERT Neural Network - EXPLAINED! Encoder in Transformer itself can learn: Relations among words (what word is most probable in a context). For instance, what word will fit in the BLANK in the context I take [BLANK] of the opportunity. Relations ...

0

Since min_count and threshold are hyperparameters, better values could be found through cross validation. Evaluate a range of values to empirically find the values that have the highest performance on a validation set.

0

You can use Transformer models such as Bert, Distilbert to build your own NLP engine. To understand how to integrate these models to interpret the input query check this article: https://ai-brewery.medium.com/simple-chatbot-using-bert-and-pytorch-part-1-2735643e0baa

0

You can use TF-IDF, TextRank, TopicRank, YAKE!, and KeyBERT for keyword extraction. Check this article: https://towardsdatascience.com/keyword-extraction-python-tf-idf-textrank-topicrank-yake-bert-7405d51cd839

0

Try the following: Check if text preprocessing is done correctly. With different sequence lengths (Ex: 1000, 750, 500 etc.) After the LSTM layer, try adding the ANN layers and check the results. Check the Transformers architecture for text classification. Reference: https://ai-brewery.medium.com/simple-chatbot-using-bert-and-pytorch-part-1-2735643e0baa

1

I guess there is no universal solution for that. But I can explain my roadmap as NLP scientist. Firstly, I try to find the most common dataset for my task (NER in your case). Then, I search for the leaderboard which shows the best papers/models for that dataset. Finally, I try to figure out which makes their models best in the leaderboard. For example, here ...

0

I had created a similar application some time back, I had extracted the features(important defining terms) from the corpus using TF-IDF and then calculated word similarity between these words with my input words and aggregated the results. You could use word embeddings like GloVe if you want to compare these words semantically.

2

Text vectorisation is a good way to have a reliable classification. You have several libraries like doc2vec that you can use together with logistic regression or dimensional reduction technique like tSNE or UMAP. https://radimrehurek.com/gensim/auto_examples/tutorials/run_doc2vec_lee.html On the other hand, you can also use libraries like BERT or TF-IDF: ...

0

You can extract a random sample from the whole data set. Start with a very small one (~1000 rows) to build a first recommendation system. Then you can increase the quantity depending on the results and the computation capabilities. If you have new data, just repeat the same process.

0

There are many options to spend this up: Get a better CPU. Distribute the process across a cluster since each document is independent. Reduce the size of the vocabulary. If only the top-n most popular words are used, it greats reduces the size of the data. Reduce the size of the embedding space. Switch to doc2vec so the document themselves are a learned ...

3

In fact, you want to translate "yellow that is glossy and sorta dark" by (170,173,11). A good way to solve this, is by using a neural machine translation model. Therefore, you can use a encoder/decoder system like many translation models, but with 3 digits as output. To achieve this, you will want to have training data with plenty of text to color ...

0

I think I have figured out how to do this: The key is to use spawn not fork, and use cupy to select GPU. import multiprocessing as mp mp.set_start_method('spawn', force=True) from joblib import Parallel, delayed from itertools import cycle import cupy import spacy from thinc.api import set_gpu_allocator, require_gpu def chunker(iterable, total_length, ...

1

Well sure it's doable: the NLP part is just speech recognition + extracting the formal command from the text, it's very similar to "virtual assistants" like Apple Siri, Amazon Alexa, Ok Google. However the hard part is to formalize all the possible commands that can be given, and then train a model to correctly map voice commands to software ...

1

Assuming your goal is to infer whether a sentence is: positive already happened, positive likely to happen, negative already happened or negative likely to happen; you end up with a 4-classes classification problem, which you need to label in advance (this would be, if feasible, the tedious-human work). After that, you can also apply word embedding layers to ...

2

It's common to see some confusion about TFIDF so thank you for asking this question :) TFIDF is not a metric, it's a weighting scheme This means that it's a way to represent a document, not to compare documents. TFIDF assumes a bag of words (BoW) representation, i.e. a document or sentence is represented as a set of words (their order doesn't matter). The ...

0

One way to approach it: Define a center tendency of the documents, a location in vector space. Then, define a distance metric (e.g., cosine, Minkowski, or Mahalanobis). Lastly, set a threshold in the distance metric that would define an outlier.

1

So, Clustering is "Unsupervised" learning : You make groups in which elements look like each-other. In Unsupervised learning, you don't have a Label that you look for. Here, your problem is to Classify text between 3 categories : Sports, Foreign, Local. Those 3 categories ARE labels : You know you have news about those 3 subjects, and want to make ...

0

There are many. A couple of the most popular: ML-Annotate - Supports binary, multi-label and multi-class labeling. TagEditor - A Windows application that uses spaCy

0

One option is start with pattern matching with established tools. For each statement, find the subject and sentiment. Here are off-the-shelf tools (no training or machine learning): WordNet a large lexical database of English words. Synset is a way to find instances are the groupings of synonymous words that express the same concept. SentiWordNet assigns to ...

0

Graph embedding gives an embedding/vector per node. That is analogous to word embedding in NLP, which gives one vector per word (often methods are quite related, e.g. word2vec vs node2vec, deepwalk etc). If you want to embed paths, that sound analogous to "sentence embedding". There are a bunch of methods you could find for that (inc RNNs etc), but ...

0

Word2vec uses a sigmoid output layer, so (unlike in softmax) each dimension of the output (corresponding to the second word in the pair) is treated completely separately. So putting through two words at the same time'' doesn't make much sense. If you completely changed things and use softmax, so that you can put multiple outputs through together, then it ...

0

Your question is valid. There are couple of known issues when trying to fit BERT-large version on small datasets (small implies a couple of 1000 training data points). The number of parameters itself is not a primary source of concern. The issues chiefly are - the use of a non-standard optimizer introduces bias in the gradient estimation; the top layers of ...

0

Assuming that the goal is to find paths which are similar with each other in the dataset, I would suggest trying to directly compare pairs of paths with an appropriate similarity/distance function. Since the order in the path is clearly relevant, I think a sequence-based measure like the Levenshtein edit distance is a good candidate. The idea would be to ...

1

"promotional" is not an inflected form of "promotion", therefore "promotion" is not the lemma of "promotional". Actually, "promotion" is a noun and "promotional" is an adjective. Maybe what you actually want to do is not lemmatisation but stemming. Note that the stem is the root of the word and, ...

-1

BERT is a stack of deep bidirectional transformer encoders that read the input sequence and generate meaning representations called embeddings. It uses multi-head attention to decide the meaning.

0

Since all the sentences length are not highly varying, you can use sentence embeddings and do the clustering on top of that. For example, Text => USE => vector[1024] => KMeans USE - Universal sentence encoders Kmeans - SKlearn Module You can adjust the number of clusters using these techniques.

0

Try: from sklearn.model_selection import train_test_split, GridSearchCV from sklearn.pipeline import Pipeline from sklearn.ensemble import RandomForestClassifier from sklearn.feature_extraction.text import TfidfVectorizer # features = ['description'] # target = 'ratingCategory' x_train, x_val, y_train, y_val = train_test_split( # train[features], ...

0

I just found myself the perfect solution. All you need to do is to write an embedding layer and a bi-Directional LSTM mine is this, input_layer = tf.keras.Input(shape = (max_len,)) embeding_layer = tf.keras.layers.Embedding(top_wordings, embeding_length, input_length= max_len) (input_layer) lstm_layer = tf.keras.layers.Bidirectional( tf.keras.layers.LSTM(15, ...

1

One way to understand how ELMo's character convolutions work is by directly inspecting the source code. There, in the forward method, you can see that the input to the network is a tensor of dimensions (batch_size, sequence_length, 50), where 50 is the maximum number of characters per word. Therefore, before passing the text to the network, it is segmented ...

0

State of the art in synonym detection tend to be ensembles of embeddings and knowledge graphs. Embeddings model tokens as dense, semantic vector representations. Knowledge graphs model the relationships between entities and are useful for adding constraints.

0

The problem you describe looks very close to standard information retrieval: given a predefined set of documents $D$ and an input string $s$, find the most similar document $d\in D$ to $s$ (alternatively find the top $n$ documents $d$ most similar to $s$). The approach you describe is good, except that in general the input string $s$ is not part of the TFIDF ...

0

Word2vec (or, e.g. GloVe) learn similar embeddings for words that occur in *similar contexts (i.e. co-occur with the same distribution of words) not words co-occurring directly.$^*$ This makes sense intuitively since words that appear together aren't necessarily similar (e.g. synonymous), but may be otherwise related, e.g. blue and sky. e.g. say film and ...

1

Extensible Markup Language (XML) is a way to encoding documents in a format that is both human-readable and machine-readable. It is also relatively simple and commonly used. It allows data and metadata to be linked. Those are the reasons that is often used for treebank data.

0

You probably do not need to use word2vec to disambiguate authors. It might be effect to use regular expressions to parse names and then do a web search. If you do want to train word2vec disambiguate authors, it would be better to embed all possible information (e.g., authors, title, journal, abstract, ...).

1

The BIO format (and its variants) is a standard format for training a sequence labeling model, in particular a Named Entity Recognition (NER) model. Sequence labeling consists in assigning a label to every token in the sequence, so at the "low level" stages of training and predicting the system must deal with the token and its label, as well as (...

1

Precision and recall are "hard" metrics. They are measure if the model's prediction is exactly the same as the target label. Often times systems like yours can use a more flexible metric such as top-5 error rate, the model is considered to have generated the correct response if the target label is one of the model’s top 5 predictions.

-1

I am a bit confused about your question because word2vec basically outputs a word vector/embedding for each word which is completely independent of the context of that word in the document. Word2vec is a very simple and nice embedding technique Google came up with that uses a neural network to make word embeddings by using either a Skip-gram or Continuous ...

0

Essentially the target vectors $\mathbf{w}$ and the context vectors $\mathbf{c}$ learn the same thing (they jointly factorise a matrix of word co-occurrence statistics and are two sides of the same coin). You can use either equivalently. However, a better result is given by taking their average... This is because the embeddings are trained so that $\mathbf{w}... 0 Just think of it as a simple binary logistic classifier. The data is word pairs$(w,c)$(positive sample) extracted from a large corpus and for each of those$k$negative samples, where a new$c$is drawn from a noise distribution. The model has two layers of parameters, no non-linear function between them, a sigmoid function on the output (not softmax). ... 1 TL;DR: A theoretical/mathematical explanation for why word2vec/GloVe embeddings of analogies appear to form parallelograms, and so can be "solved" by adding/subtracting embeddings, is given here, as summarised in this blog. More explanation of w2v is given here. The dimensions of word2vec (or GloVe, etc) word embeddings are not directly ... 1 you can start by using torchscript, it may require changing ur whole code, and switching to transformers( by loading the backbone of the model and the last layers) so basically u get out from GIL interpreter, coz it does not support multithreading. by with torchscript u can run ur model in c++ env, there's also onnx which I believe it enhances performance. ... 0 Your understanding is not correct. You go from a$B \times M \times D$tensor to a$B \times M \times V$tensor (i.e. the logits). As you can see, in the final tensor we have M vectors of dimension V (one vector per token), not just a single vector. To obtain the$B \times M \times V\$, you just perform a matrix multiplication. This applies to Transformers, ...

3

You are currently using the fit_transform method on both your training dataset as well as your test set. This is incorrect since you should not fit the model on your test set as (depending on the model used) this would be overfitting, and it can give issues with dataset shapes when creating new columns based on the values in the data (count vectorizer, ...

1

Since you mention deep learning, one option is to embedded the documents and then cluster the documents. Each cluster could be labeled as "Good" or "Not Good". The labeling could be done by hand or automatically by voting with existing labels (e.g., if a majority of the documents are "Good" then the entire cluster is "Good&...

0

It depends on the type of ranking that you want to achieve, for example if the unlabeled scraped data can be ranked by sentiment, you can use Transfer Learning models to give each document a sentiment score which will serve as a rank if you return the sentiment score probability instead of having "positive" and "negative" tags. Transfer ...

Top 50 recent answers are included