Exam AWS Certified Machine Learning - Specialty topic 1 question 27 discussion

B C F should be correct.

I will select B, C, F 1- Apply words stemming and lemmatization 2- Remove Stop words 3- Tokensize the sentences https://towardsdatascience.com/nlp-extracting-the-main-topics-from-your-dataset-using-lda-in-minutes-21486f5aa925

B. Normalize all words by making the sentence lowercase: Word2Vec treats words as distinct entities. If you don't convert everything to lowercase, "The" and "the" will be considered different words, which is generally not what you want. Lowercasing ensures consistency. D. Correct the typography on "quck" to "quick": Misspellings need to be corrected. Word2Vec learns embeddings based on the words it encounters. If "quck" remains, it will be treated as a separate word from "quick," and you'll lose the relationship between them. Correcting typos is crucial for data quality. F. Tokenize the sentence into words: Tokenization is the process of breaking down the sentence into individual words (or tokens). Word2Vec operates on individual words, so you need to split the sentence into its constituent parts. This is a fundamental step in NLP.

While C - is debatable - not always necessary to remove stop words in Word2Vec - as sometimes the stop words do provide context ==================== For Word2Vec training, data preprocessing is essential to ensure that words are correctly represented, consistent, and free from unnecessary noise. The key steps are: Lowercasing the text (B) Word embeddings treat "FOX" and "fox" as different words. To avoid redundancy, lowercasing the text ensures consistency. Correcting typos (D) "quck" should be corrected to "quick" to prevent incorrect word representations in Word2Vec. Misspelled words can create meaningless embeddings. Tokenizing the sentence into words (F) Word2Vec operates at the word level, so breaking the sentence into individual tokens (words) is necessary.

A. NO - word2vec works on raw data B. YES - case here is not significant C. YES - will help reduce dimensionality D. NO - word2vec will do it by itself E. NO - One-hot encoding is for classification F. YES - word2vec takes tokens as input

Data need to be tokenized and cleaned!

B, C F is the correct

BCF correct. D is not correct (Pay attention to “in a repeatable manner” in the question.)

B/C/F. D should not be performed because spell check is a subjective thing. You don't know for sure what the word was supposed to be if you have a typo.

I saw this exact question on "whizlabs" practice exam and correct options were B/C/F

https://towardsdatascience.com/an-implementation-guide-to-word2vec-using-numpy-and-google-sheets-13445eebd281 Data Preparation — Define corpus, clean, normalise and tokenise words To begin, we start with the following corpus: “natural language processing and machine learning is fun and exciting” For simplicity, we have chosen a sentence without punctuation and capitalization. Also, we did not remove stop words “and” and “is”. In reality, text data are unstructured and can be “dirty”. Cleaning them will involve steps such as o removing stop words, o removing punctuations, o convert text to lowercase (actually depends on your use-case), o replacing digits, etc. o After preprocessing, we then move on to tokenising the corpus Answer: B, C, F

Correct answers are B, C and F

The correct answer is B, C and F A: POS tagging has nothing to do with word2vec D: fixing "quck" to "quick" only works for that specific word F: word2vec can use CBOW or skipgram, so no need to have one-hot decoding here

BCF is correct

B, C F correct

B, C, and F are correct answers. I have done this question many times in many practice tests.

B, C, F are my choice. D is also possible but not as widely used as others.