This project aims to extract keywords from a text document. All text documents are written to comment, citate, confirm or deny an idea. Each of these documents has unique ideas that discrimate them from the others. Each document contains a set of words that determine its general idea that signals the readers about its content. These words or phrases are called keywords. As we know, keywords are the core of a text which can simply demonstrate the main content of a text in a few words without reading the whole text. In this project, after entering the text, first, Stop words are removed and other words will gain weights according their repetition in a text. If a word repeated several times in a text, it could be the expected keyword. In the next step, by choosing how many keywords you want to extract, keywords are shown to and you can evaluate this system. This projects is written for both Persian and English language, so you can use it to extract keywords in both...
With the growth of unstructured text data over the Internet, which is mainly the result of the human interaction in web2.0 and social networks, finding a way to automatically process and extract knowledge from this data seems indispensable. Despite unstructured format this data contain valuable knowledge which can be extracted using knowledge discovery and machine learning techniques. There has been great progress in natural language processing task such as Sentiment analysis, Opinion mining, Topic identification, Automatic machine translation, Name entity recognition, Part of speech tagging, Parsing, Information extraction, Question answering, Paraphrase detection, etc. In most of NLP tasks we first develop an algorithm and then convert our data to be prepared to feed into that algorithm. This is called feature engineering which is very time consuming. Mainly, words are considered as features in text data. But there are two shortcomings in this method: First word order may be lost and second is the sparsity of feature vector which affect training time. The aim of this project is to find a way to automatically do feature extraction from text data in Persian. We found deep learning as a way to deal with this problem. Neural network with more than one layer is called deep network. In this method each word is described with a numerical vector, which is called distributed representation or word vector. This representation contains semantic and syntactic information about words. Word concatenation represent sentences. If we can describe words with such vectors the sentence could be too. The range of this combination include simple mathematic operator like vector addition or multiplication, to recurrent neural network and recursive...
Home > Projects Automatic speech recognition (ASR) is the translation of spoken words into text. Speech recognition has many applications such as virtual speech assistants (e.g., Apple’s Siri, Google Now, and Microsoft’s Cortana), speech-to-speech translation, voice dictation and etc. As shown in Fig. 1, an ASR system has five main components: signal processing and feature extraction, acoustic model (AM), language model (LM), lexicon and hypothesis search. The signal processing and feature extraction component takes the audio signal as the input, enhances the speech by removing noises and extracts feature vectors. The acoustic model integrates knowledge about acoustics and phonetics, takes the features as its input, and recognize phonemes. Language model contains information about structure of the language. Lexicon includes all words that audio signal can be mapped to them. The hypothesis search component combines AM and LM scores and outputs the word sequence with the highest score as the recognition result. There are several methods for creating acoustic model, such as hidden markov model (HMM) [5] and artificial neural network (ANN) [1]. Audio signals as a sequential data, current input depends on previous inputs. Recurrent neural networks (RNNs) [1] benefits for their ability to learn sequential data. But for standard RNN architectures, the range of context that can be in practice accessed is quite limited. This problem is often referred to in the literature as the vanishing gradient problem. In 1997, after introducing long short term memory (LSTM) [1-3] neural network, the problem of limitation in processing sequential data resolved. An LSTM network is the same as a standard RNN, except that units in the hidden layer are replaced by...
Home > Projects In this project, a Persian Question Answering (QA) system is created to ease the access to information resources for doctors, health providers and users. To this aim, a set of Persian documents related to drugs and diseases are collected. The processing of the structured documents improves the performance of the QA system that’s why all documents were converted into semi-structured documents. The developed system consists of three main units: question processing document retrieval answer extraction The question processing unit, as in the most important module, consists of four components that sequentially extract keywords/queries. These components use a dictionary of drugs/diseases names and keywords/queries. This process is shown in the following figure. If a module fails to extract keywords from the question, based on the condition of the question, another component would make the extraction process instead. The first part of the QA system is question processing module. The main component of the question processing module includes Question Classifier, N-gram Tokenizer, Patterns Matching and Advanced Tokenizer. In this architecture, the question asked by the user, is normalized and then the drug name or disease related to the question is extracted through Name Entity (NE) Dictionary. If this specified name is extracted from the question, the question would be sent to Question Classifier component for the extraction of the phrases that indicate the meaning of the question. Finally, by using the concept of the dictionary, the keywords would be extracted and the phrases would be mapped to the dictionary keywords. On the other hand, if the Question Classifier fails to extract any keywords, the question would be...
Home > Projects Human society functions by communication between individuals. Language in both its written and spoken forms underpin all aspects of human interactions. The spoken language is the most fundamental as this is how individuals communicate with one another using only the human vocal apparatus. Since spoken language is one of the easiest measures to acquire (all your need is a microphone), is used in a variety of transaction applications (e.g. telephone banking), and has the potential for security by surveillance, it comes as no surprise that speaker recognition is one of the key research areas in signal processing and pattern recognition. Deep Belief Networks (DBNs) have become a popular research area in machine learning and pattern recognition. In recent years, deep learning techniques have been successfully applied to the modeling of speech signals, such as speech recognition, acoustic modeling, speaker and language recognition, spectrogram coding, voice activity detection, acoustic-articulatory inversion mapping, 3D object recognition, intelligent video surveillance and image recognition. DBNs, use a powerful strategy of unsupervised training and multiple layers that provides parameter-efficient and accurate acoustic modeling. DBNs have been successfully used in speech recognition for modeling the posterior probability of state given a feature vector. Feature vectors are typically standard frame based acoustic representations (e.g., MFCCs) that are usually stacked across multiple frames. The DBN performs a nonlinear transformation of the input features, and produces the probability that an output unit is active, given a wide context of input frames. The basic process for pre-training a DBN is based upon stacking RBMs. RBMs are an undirected graphical model with visible and hidden units with...
