Tendencies of development science and practice 330 algorithms for interpreting word vectors
Download 304.03 Kb. Pdf ko'rish
|
TENDENCIES-OF-DEVELOPMENT-SCIENCE-AND-PRACTICE-331-339 (1)
- Bu sahifa navigatsiya:
- 2.Word2vec.
Advantages
• Simple architecture: feed-forward , 1 input, 1 hidden layer, 1 output. • The model quickly learns and generates embeddings (even your own). • Embeddings are endowed with meaning, controversial points can be deciphered. • The methodology can be extended to many other areas/problems (eg Lda2vec ). Flaws • While the co-occurrence matrix provides global information, GloVe remains trained at the word level and provides little data about the sentence and the context in which the word is used. • Handles unknown and rare words poorly. Source : Jeffrey Pennington, Richard Socher, and Christopher D. Manning, GloVe : Global Vectors for Word Representation (2014), Empirical Methods in Natural Language Processing 2.Word2vec. In 2013, Tomas Mikolov of Google proposed a more efficient learning model for word vector representations, Word2vec. The method was based on the assumption that words that are often found in the same contexts have similar meanings. The changes were simple - removing the hidden layer and approximating (simplifying) the target - but became a turning point in the development of NLP language models. Instead of a continuous bag of words algorithm, the Word2Vec model uses Skip- gram (skip phrase). The purpose of this model is exactly the opposite of the previous model - to predict the surrounding words based on the central one. Continuous bag of words and Skip-gram architectures. Skip Gram A "context window" is formed - a sequence of k words in the text. One of these words is skipped, and the neural network tries to predict it. Thus, words that often occur in a similar context will have similar vectors. TENDENCIES OF DEVELOPMENT SCIENCE AND PRACTICE 336 Negative Sampling Many words in the texts do not occur together, so the model performs a lot of unnecessary calculations. Calculating softmax is a computationally expensive operation. The Negative Sampling approach allows you to maximize the probability of meeting the desired word in a context that is typical for it, and minimize it in a rare / atypical context. Vector magic. The Word2vec model amazed researchers with its "interpretability". Learning on large corpora of texts allows you to determine deep relationships between word forms, for example, gender. If the vector corresponding to the word Man (Man) is subtracted from the vector Woman (Woman), the result will be very similar to the difference between the vectors King (King) and Queen (Queen). At one time, such a relationship between words and their vectors seemed almost magic. You can find some more fun examples of vector arithmetic in the article What life becomes without love. Despite the huge contribution that the model made to NLP, now it is almost not used - worthy heirs have replaced it. Finished Models The pretrained model is readily available online. It can be imported into a Python project using the gensim library. Advantages ❖ Simple architecture: feed-forward, 1 input, 1 hidden layer, 1 output. ❖ The model quickly learns and generates embeddings (even your own). ❖ Embeddings are endowed with meaning, controversial points can be deciphered. TENDENCIES OF DEVELOPMENT SCIENCE AND PRACTICE 337 ❖ The methodology can be extended to many other areas/problems (eg Lda2vec). • Flaws • Word-level learning: No information about the sentence or the context in which the word is used. • Co-occurrence is ignored. The model does not take into account the fact that a word can have a different meaning depending on the context of use. This is the main reason why GloVe is generally preferred over Word2Vec. FUTURE WORK NLP’s future will be redefined as it faces new technological challenges and a push from the market to create more user friendly systems. Market’s influence is prompting fiercer competition among existing NLP based companies. It is also pushing NLP more towards Open Source Development. If the NLP community embraces Open Source Development, it will make NLP systems less proprietary and therefore less expensive. The systems will also be built as easily replaceable components, which take less time to build and more userfriendly [9]. Chatterbots – although they exist already, new generations of them are being constantly developed. Chatterbots use natural language processing to simulate conversations with users. Web sites are beginning to install chatterbots as Web guides and customer service agents. CONCLUSION While NLP is a relatively recent area of research and application, as compared to other information technology approaches, there have been sufficient successes to date that suggest that NLP-based information access technologies will continue to be a major area of research and development in information systems now and far into the future. The importance of NLP in processing the input text to be synthesized is reflected. The naturalness of the speech utterances produced by the signal-processing modules are tightly bound to the performance of the previous text-processing modules. Download 304.03 Kb. Do'stlaringiz bilan baham: |
Ma'lumotlar bazasi mualliflik huquqi bilan himoyalangan ©fayllar.org 2024
ma'muriyatiga murojaat qiling
ma'muriyatiga murojaat qiling