Word Embeddings, word2vec, Sentiment Analysis up to BERT скачать в хорошем качестве

Word Embeddings, word2vec, Sentiment Analysis up to BERT

The code for the video: https://github.com/sirajzade/learning... More information about me: https://sirajzade.github.io 0:00 Introduction 1:00 Numerical Representation for Text 2:08 Word2vec demo 3:52 Word Vectors and Word Emdedding 5:37 Skip Gram, CBOW; Distributional Semantics 10:35 Sentiment Analysis Use Case Since Word2vec algorithm was published in 2013, Word Embeddings are on everyone’s lips. In this video I am going to explain what it is and how it can be used with the concrete code examples. With this fast and handy technology one can solve many NLP problems easily. Its comprehension helps in the understanding of other deep learning technologies in NLP. The newest technologies like BERT for example use internally also embeddings although its structure is different. Hi, my name is Joshgun Sirajzade and I am a PostDoc Researcher at University of Luxembourg. One of the most foundational questions of NLP is how to create numbers out of the words. This is also true for text mining, machine learning or any other field dealing with text computationally. The phenomenon is also called creating a numerical representation for words or for text. One of the ways of doing it would be counting the occurrence of words in a document. The most important question here however is how to create a number which represents a particular word best in one or the other way. The goal is not only to be able to make sense of words, but also to be able to compare them to each other. This can be done on a word level or even on a sentence or document level. In fact, the whole field of NLP is about processing and making sense of text data or human speech. If you are interested in getting an idea about all application fields of NLP, watch my video about it. So, let me first show you what word2vec can do in a code example and then we can discuss how the algorithm works. Word2vec has many implementations, in this video we are going to use the implementation from genism library. In the first line we import the genism library and word2vec algorithm from it. In genism library you can download pre trained models or some text examples to train own your models. This is what we do from line 4 to 7. First, we download an example text, then we train a model. What a model is I will explain later in this video. In order save time computationally and not to train your model again and again, you can save it locally and load it when you need it. So, one powerful and very useful functoriality here is the ability to give any word and get the most similar words to it. In gensim word2vec this is done with the function most similar. In our example, if we look at the word “president” we get similar words to it like “governor”, “chairman”, “senator” and even “chancellor”. For the word “city” we get “town”, “suburbs”, “downtown”, “village” and many others. For “coffee” we get words like “sugar”, “cocoa” and many other food and drink related words. This ability of word2vec clarifies best what it does. Please keep in mind that these words are not synonyms nor antonyms, because many people mistaken them for those. In the case of word2vec we speak about these words being semantically similar or related to each other without defining it further in detail. Technically, similar here means that vectors of these words occur or appear close to each other in the vector space. Because, as the name suggests, word2vec yields vectors from words a.k.a. numerical representations for words. A vector is a mathematical object, but you can imagine it as a number representing a word. However, because words have semantically complex relationships, usually one number per word is not enough to represent it, that is why we use many numbers per word, which is mathematically called a vector. I made a separate video about Vector Space model and Document Term Matrix, you can watch it, in order to get a better intuition how vectors from text are created. In that video I mention that the vector of a word represents the documents it occurred in, this is very easy to see, especially in the one hot vectors. In this way one can compare words to each other. Words which occur in the same document will have similar vectors. Word2vec leverages the same idea, however, instead of using documents it uses a so-called window which would be a smaller unit than documents and sentences. Usually, the window takes the context of the words, forming it from 5 words before the word and 5 after it. However, this number can be changed manually or automatically.