I am working on a sentiment analysis solution with BERT to analyze tweets in german. My training dataset of is a class of 1000 tweets, which have been manually annotated into the classes neutral, positive and negative.
The dataset with 10.000 tweets is quite unevenly distributed:
approx.
3000 positive
2000 negative
5000 neutral
the tweets contain formulations with #names, https links, numbers, punctuation marks, smileys like :3 :D :) etc..
The interesting thing is, if I remove them with the following code during Data Cleaning, the F1 score gets worse. Only the removal of https links (if I do it alone) leads to a small improvement.
# removing the punctuation and numbers
def remove_punct(text):
text = re.sub(r'http\S+', '', text) # removing links
text = re.sub(r'#\S+', '', text) # removing referencing on usernames with #
text = re.sub(r':\S+', '', text) # removing smileys with : (like :),:D,:( etc)
text = "".join([char for char in text if char not in string.punctuation])
text = re.sub('[0-9]+', '', text)
return text
data['Tweet_clean'] = data['Tweet'].apply(lambda x: remove_punct(x)) # extending the dataset with the column tweet_clean
data.head(40)
also steps like stop words removal or lemmitazation lead more to a deterioration. Is this because I do something wrong or can the model BERT actually handle such values?
A second question is:
I found other records that were also manually annotated, but these are not tweets and the structure of the sentences and language use is different. Would you still recommend to add these records to my original?
There are about 3000 records in German.
My last question:
Should I reduce the class sizes to the size of the smallest unit and thus balance?
BERT can handle punctuation, smileys etc. Of course, smileys contribute a lot to sentiment analysis. So, don't remove them. Next, it would be fair to replace #mentions and links with some special tokens, because the model will probably never see them again in the future.
If your model is designed for tweets, I suggest that you fine-tune BERT with additional corpus, and after fine-tune with Twitter corpus. Or do it simultaneously. More training samples is generally better.
No, it is better to use class weights instead of downsampling.
Based on this paper (By Adam Ek, Jean-Philippe Bernardy and Stergios Chatzikyriakidis), BERT models outperform BiLSTM in terms of better generalizing to punctuations. Looking at the experiments' results in the paper, I say keep the punctuations.
I couln't find anything solid for smiley faces; However, after doing some experiments with the HuggingFace API, I didn't notice much difference with/without smiley faces.
Related
I want to predict the occurrences of the word "repeat" in a speech as well as the word's approximate duration. For this task, I'm planning to build a Deep Learning model. I've around 50 positive as well as 50 negative utterances (I couldn't collect more).
Initially I've searched for any pretrained models for keyword spotting, but I couldn't get a good one.
Then I tried Speech Recognition models (Deep Speech), but it couldn't predict the exact repeat words as my data follows Indian accent. Also, I've thought that going for ASR models for this task would be a over-killing one.
Now, I've split the entire audio into chunk of 1 secs with 50% overlapping and tried a binary audio classification in each chunk that is whether the chunk has the word "repeat" or not. For building the classification model, I calculated the MFCC features and build a sequence model on the top of it. Nothing seems to work for me.
If anyone already worked with this kind of task, please provide me with a correct method/resources to build a DL model for this task. Thanks in advance!
Is deep learning model supports multi-label classification problem or any other algorithms in H2O?
Orginal Response Variable -Tags:
apps, email, mail
finance,freelancers,contractors,zen99
genomes
gogovan
brazil,china,cloudflare
hauling,service,moving
ferguson,crowdfunding,beacon
cms,naytev
y,combinator
in,store,
conversion,logic,ad,attribution
After mapping them on the keys of the dictionary:
Then
Response variable look like this:
[74]
[156, 89]
[153, 13, 133, 40]
[150]
[474, 277, 113]
[181, 117]
[15, 87, 8, 11]
Thanks
No, H2O only contains algorithms that learn to predict a single response variable at a time. You could turn each unique combination into a single class and train a multi-class model that way, or predict each class with a separate model.
Any algorithm that creates a model that gives you "finance,freelancers,contractors,zen99" for one set of inputs, and "cms,naytev" for another set of inputs is horribly over-fitted. You need to take a step back and think about what your actual question is.
But in lieu of that, here is one idea: train some word embeddings (or use some pre-trained ones) on your answer words. You could then average the vectors for each set of values, and hope this gives you a good numeric representation of the "topic". You then need to turn your, say, 100 dimensional averaged word vector into a single number (PCA comes to mind). And now you have a single number that you can give to a machine learning algorithm, and that it can predict.
You still have a problem: having predicted a number, how do you turn that number into a 100-dim vector, and from there in to a topic, and from there into topic words? Tricky, but maybe not impossible.
(As an aside, if you turn the above "single number" into a factor, and have the machine learning model do a categorization, to predicting the most similar topic to those it has seen before... you've basically gone full circle and will get a model identical to the one you started with that has too many classes.)
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 2 years ago.
Improve this question
I'm going through this tutorial on RNNs/LSTMs and I'm having quite a hard time understanding stateful LSTMs. My questions are as follows :
1. Training batching size
In the Keras docs on RNNs, I found out that the hidden state of the sample in i-th position within the batch will be fed as input hidden state for the sample in i-th position in the next batch. Does that mean that if we want to pass the hidden state from sample to sample we have to use batches of size 1 and therefore perform online gradient descent? Is there a way to pass the hidden state within a batch of size >1 and perform gradient descent on that batch ?
2. One-Char Mapping Problems
In the tutorial's paragraph 'Stateful LSTM for a One-Char to One-Char Mapping' were given a code that uses batch_size = 1 and stateful = True to learn to predict the next letter of the alphabet given a letter of the alphabet. In the last part of the code (line 53 to the end of the complete code), the model is tested starting with a random letter ('K') and predicts 'B' then given 'B' it predicts 'C', etc. It seems to work well except for 'K'. However, I tried the following tweak to the code (last part too, I kept lines 52 and above):
# demonstrate a random starting point
letter1 = "M"
seed1 = [char_to_int[letter1]]
x = numpy.reshape(seed, (1, len(seed), 1))
x = x / float(len(alphabet))
prediction = model.predict(x, verbose=0)
index = numpy.argmax(prediction)
print(int_to_char[seed1[0]], "->", int_to_char[index])
letter2 = "E"
seed2 = [char_to_int[letter2]]
seed = seed2
print("New start: ", letter1, letter2)
for i in range(0, 5):
x = numpy.reshape(seed, (1, len(seed), 1))
x = x / float(len(alphabet))
prediction = model.predict(x, verbose=0)
index = numpy.argmax(prediction)
print(int_to_char[seed[0]], "->", int_to_char[index])
seed = [index]
model.reset_states()
and these outputs:
M -> B
New start: M E
E -> C
C -> D
D -> E
E -> F
It looks like the LSTM did not learn the alphabet but just the positions of the letters, and that regardless of the first letter we feed in, the LSTM will always predict B since it's the second letter, then C and so on.
Therefore, how does keeping the previous hidden state as initial hidden state for the current hidden state help us with the learning given that during test if we start with the letter 'K' for example, letters A to J will not have been fed in before and the initial hidden state won't be the same as during training ?
3. Training an LSTM on a book for sentence generation
I want to train my LSTM on a whole book to learn how to generate sentences and perhaps learn the authors style too, how can I naturally train my LSTM on that text (input the whole text and let the LSTM figure out the dependencies between the words) instead of having to 'artificially' create batches of sentences from that book myself to train my LSTM on? I believe I should use stateful LSTMs could help but I'm not sure how.
Having a stateful LSTM in Keras means that a Keras variable will be used to store and update the state, and in fact you could check the value of the state vector(s) at any time (that is, until you call reset_states()). A non-stateful model, on the other hand, will use an initial zero state every time it processes a batch, so it is as if you always called reset_states() after train_on_batch, test_on_batch and predict_on_batch. The explanation about the state being reused for the next batch on stateful models is just about that difference with non-stateful; of course the state will always flow within each sequence in the batch and you do not need to have batches of size 1 for that to happen. I see two scenarios where stateful models are useful:
You want to train on split sequences of data because these are very long and it would not be practical to train on their whole length.
On prediction time, you want to retrieve the output for each time point in the sequence, not just at the end (either because you want to feed it back into the network or because your application needs it). I personally do that in the models that I export for later integration (which are "copies" of the training model with batch size of 1).
I agree that the example of an RNN for the alphabet does not really seem very useful in practice; it will only work when you start with the letter A. If you want to learn to reproduce the alphabet starting at any letter, you would need to train the network with that kind of examples (subsequences or rotations of the alphabet). But I think a regular feed-forward network could learn to predict the next letter of the alphabet training on pairs like (A, B), (B, C), etc. I think the example is meant for demonstrative purposes more than anything else.
You may have probably already read it, but the popular post The Unreasonable Effectiveness of Recurrent Neural Networks shows some interesting results along the lines of what you want to do (although it does not really dive into implementation specifics). I don't have personal experience training RNN with textual data, but there is a number of approaches you can research. You can build character-based models (like the ones in the post), where your input and receive one character at a time. A more advanced approach is to do some preprocessing on the texts and transform them into sequences of numbers; Keras includes some text preprocessing functions to do that. Having one single number as feature space is probably not going to work all that well, so you could simply turn each word into a vector with one-hot encoding or, more interestingly, have the network learn the best vector representation for each for, which is what they call en embedding. You can go even further with the preprocessing and look into something like NLTK, specially if you want to remove stop words, punctuation and things like that. Finally, if you have sequences of different sizes (e.g. you are using full texts instead of excerpts of a fixed size, which may or may not be important for you) you will need to be a bit more careful and use masking and/or sample weighting. Depending on the exact problem, you can set up the training accordingly. If you want to learn to generate similar text, the "Y" would be the similar to the "X" (one-hot encoded), only shifted by one (or more) positions (in this case you may need to use return_sequences=True and TimeDistributed layers). If you want to determine the autor, your output could be a softmax Dense layer.
Hope that helps.
I am new to keras and despite reading the documentation and the examples folder in keras, I'm still struggling with how to fit everything together.
In particular, I want to start with a simple task: I have a sequence of tokens, where each token has exactly one label. I have a lot training data like this - practically infinite, as I can generate more (token, label) training pairs as needed.
I want to build a network to predict labels given tokens. The number of tokens must always be the same as the number of labels (one token = one label).
And I want this to be based on all surrounding tokens, say within the same line or sentence or window -- not just on the preceding tokens.
How far I got on my own:
created the training numpy vectors, where I converted each sentence into a token-vector and label-vector (of same length), using a token-to-int and label-to-int mappings
wrote a model using categorical_crossentropy and one LSTM layer, based on https://github.com/fchollet/keras/blob/master/examples/lstm_text_generation.py.
Now struggling with:
All the input_dim and input_shape parameters... since each sentence has a different length (different number of tokens and labels in it), what should I put as input_dim for the input layer?
How to tell the network to use the entire token sentence for prediction, not just one token? How to predict a whole sequence of labels given a sequence of tokens, rather than just label based on previous tokens?
Does splitting the text into sentences or windows make any sense? Or can I just pass a vector for the entire text as a single sequence? What is a "sequence"?
What are "time slices" and "time steps"? The documentation keeps mentioning that and I have no idea how that relates to my problem. What is "time" in keras?
Basically I have trouble connecting the concepts from the documentation like "time" or "sequence" to my problem. Issues like Keras#40 didn't make me any wiser.
Pointing to relevant examples on the web or code samples would be much appreciated. Not looking for academic articles.
Thanks!
If you have sequences of different length you can either pad them or use a stateful RNN implementation in which the activations are saved between batches. The former is the easiest and most used.
If you want to use future information when using RNNs you want to use a bidirectional model where you concatenate two RNN's moving in opposite directions. RNN will use a representation of all previous information when e.g. predicting.
If you have very long sentences it might be useful to sample a random sub-sequence and train on that. Fx 100 characters. This also helps with overfitting.
Time steps are your tokens. A sentence is a sequence of characters/tokens.
I've written an example of how I understand your problem but it's not tested so it might not run. Instead of using integers to represent your data I suggest one-hot encoding if it is possible and then use binary_crossentropy instead of mse.
from keras.models import Model
from keras.layers import Input, LSTM, TimeDistributed
from keras.preprocessing import sequence
# Make sure all sequences are of same length
X_train = sequence.pad_sequences(X_train, maxlen=maxlen)
# The input shape is your sequence length and your token embedding size (which is 1)
inputs = Input(shape=(maxlen, 1))
# Build a bidirectional RNN
lstm_forward = LSTM(128)(inputs)
lstm_backward = LSTM(128, go_backwards=True)(inputs)
bidirectional_lstm = merge([lstm_forward, lstm_backward], mode='concat', concat_axis=2)
# Output each timestep into a fully connected layer with linear
# output to map to an integer
sequence_output = TimeDistributed(Dense(1, activation='linear'))(bidirectional_lstm)
# Dense(n_classes, activation='sigmoid') if you want to classify
model = Model(inputs, sequence_output)
model.compile('adam', 'mse')
model.fit(X_train, y_train)
Suppose I want to match address records (or person names or whatever) against each other to merge records that are most likely referring to the same address. Basically, I guess I would like to calculate some kind of correlation between the text values and merge the records if this value is over a certain threshold.
Example:
"West Lawnmower Drive 54 A" is probably the same as "W. Lawn Mower Dr. 54A" but different from "East Lawnmower Drive 54 A".
How would you approach this problem? Would it be necessary to have some kind of context-based dictionary that knows, in the address case, that "W", "W." and "West" are the same? What about misspellings ("mover" instead of "mower" etc)?
I think this is a tricky one - perhaps there are some well-known algorithms out there?
A good baseline, probably an impractical one in terms of its relatively high computational cost and more importantly its production of many false positive, would be generic string distance algorithms such as
Edit distance (aka Levenshtein distance)
Ratcliff/Obershelp
Depending on the level of accuracy required (which, BTW, should be specified both in terms of its recall and precision, i.e. generally expressing whether it is more important to miss a correlation than to falsely identify one), a home-grown process based on [some of] the following heuristics and ideas could do the trick:
tokenize the input, i.e. see the input as an array of words rather than a string
tokenization should also keep the line number info
normalize the input with the use of a short dictionary of common substituions (such as "dr" at the end of a line = "drive", "Jack" = "John", "Bill" = "William"..., "W." at the begining of a line is "West" etc.
Identify (a bit like tagging, as in POS tagging) the nature of some entities (for example ZIP Code, and Extended ZIP code, and also city
Identify (lookup) some of these entities (for example a relative short database table can include all the Cities / town in the targeted area
Identify (lookup) some domain-related entities (if all/many of the address deal with say folks in the legal profession, a lookup of law firm names or of federal buildings may be of help.
Generally, put more weight on tokens that come from the last line of the address
Put more (or less) weight on tokens with a particular entity type (ex: "Drive", "Street", "Court" should with much less than the tokens which precede them.
Consider a modified SOUNDEX algorithm to help with normalization of
With the above in mind, implement a rule-based evaluator. Tentatively, the rules could be implemented as visitors to a tree/array-like structure where the input is parsed initially (Visitor design pattern).
The advantage of the rule-based framework, is that each heuristic is in its own function and rules can be prioritized, i.e. placing some rules early in the chain, allowing to abort the evaluation early, with some strong heuristics (eg: different City => Correlation = 0, level of confidence = 95% etc...).
An important consideration with search for correlations is the need to a priori compare every single item (here address) with every other item, hence requiring as many as 1/2 n^2 item-level comparisons. Because of this, it may be useful to store the reference items in a way where they are pre-processed (parsed, normalized...) and also to maybe have a digest/key of sort that can be used as [very rough] indicator of a possible correlation (for example a key made of the 5 digit ZIP-Code followed by the SOUNDEX value of the "primary" name).
I would look at producing a similarity comparison metric that, given two objects (strings perhaps), returns "distance" between them.
If you fulfil the following criteria then it helps:
distance between an object and
itself is zero. (reflexive)
distance from a to b is the same in
both directions (transitive)
distance from a to c is not more
than distance from a to b plus
distance from a to c. (triangle
rule)
If your metric obeys these they you can arrange your objects in metric space which means you can run queries like:
Which other object is most like
this one
Give me the 5 objects
most like this one.
There's a good book about it here. Once you've set up the infrastructure for hosting objects and running the queries you can simply plug in different comparison algorithms, compare their performance and then tune them.
I did this for geographic data at university and it was quite fun trying to tune the comparison algorithms.
I'm sure you could come up with something more advanced but you could start with something simple like reducing the address line to the digits and the first letter of each word and then compare the result of that using a longest common subsequence algorithm.
Hope that helps in some way.
You can use Levenshtein edit distance to find strings that differ by only a few characters. BK Trees can help speed up the matching process.
Disclaimer: I don't know any algorithm that does that, but would really be interested in knowing one if it exists. This answer is a naive attempt of trying to solve the problem, with no previous knowledge whatsoever. Comments welcome, please don't laugh too laud.
If you try doing it by hand, I would suggest applying some kind of "normalization" to your strings : lowercase them, remove punctuation, maybe replace common abbreviations with the full words (Dr. => drive, St => street, etc...).
Then, you can try different alignments between the two strings you compare, and compute the correlation by averaging the absolute differences between corresponding letters (eg a = 1, b = 2, etc.. and corr(a, b) = |a - b| = 1) :
west lawnmover drive
w lawnmower street
Thus, even if some letters are different, the correlation would be high. Then, simply keep the maximal correlation you found, and decide that their are the same if the correlation is above a given threshold.
When I had to modify a proprietary program doing this, back in the early 90s, it took many thousands of lines of code in multiple modules, built up over years of experience. Modern machine-learning techniques ought to make it easier, and perhaps you don't need to perform as well (it was my employer's bread and butter).
So if you're talking about merging lists of actual mailing addresses, I'd do it by outsourcing if I can.
The USPS had some tests to measure quality of address standardization programs. I don't remember anything about how that worked, but you might check if they still do it -- maybe you can get some good training data.