Search

In this research paper, we analyzed what moments and activities make people happy, based on a collection of happy moments. We are focusing on specific happy moments from a collection of text responses that people have shared through the crowd-sourcing platform: Amazon Mechanical Turk (MTurk). Using crowd-sourcing to collect our data allows us to advance our understanding of the cause of happiness, by focusing on words and real human experiences. Workers of MTurk were asked to reflect on what makes them happy in a given period and share three specific moments in complete sentences. Through text-based analysis, we will look to see what other components have a role in making a specific event happy and further analyze how we can classify such words. Also, we dive deeper into specific subcategories of classifiers in an attempt to form insights about their happiness level based on specific factors. With the goal to extract features from the text in HappyDB, in this study we used the bag of words approach. Through doing so, our results were successful at predicting the happiness category, concerning both accuracy and context. Our models were able to accomplish the goal of understanding a happy moment and fit such a moment into one of the seven ground truth happiness categories we set at the beginning of this study. We finished the article with the ethical perspective of such research works and related social implications.

Background Accurately recognizing rare diseases based on symptom description is an important task in patient triage, early risk stratification, and target therapies. However, due to the very nature of rare diseases, the lack of historical data poses a great challenge to machine learning-based approaches. On the other hand, medical knowledge in automatically constructed knowledge graphs (KGs) has the potential to compensate the lack of labeled training examples. This work aims to develop a rare disease classification algorithm that makes effective use of a knowledge graph, even when the graph is imperfect. Method We develop a text classification algorithm that represents a document as a combination of a “bag of words” and a “bag of knowledge terms,” where a “knowledge term” is a term shared between the document and the subgraph of KG relevant to the disease classification task. We use two Chinese disease diagnosis corpora to evaluate the algorithm. The first one, HaoDaiFu, contains 51,374 chief complaints categorized into 805 diseases. The second data set, ChinaRe, contains 86,663 patient descriptions categorized into 44 disease categories. Results On the two evaluation data sets, the proposed algorithm delivers robust performance and outperforms a wide range of baselines, including resampling, deep learning, and feature selection approaches. Both classification-based metric (macro-averaged F1 score) and ranking-based metric (mean reciprocal rank) are used in evaluation. Conclusion Medical knowledge in large-scale knowledge graphs can be effectively leveraged to improve rare diseases classification models, even when the knowledge graph is incomplete.

In the past fifteen years, the grow of using Bag of Words (BoW) method in the field of computer vision is visibly observed. In addition,-for the text classification and texture recognition, it can also be used in classification of images, videos, robot localization, etc. It is one of the most common methods for the categorization of text and objects. In text classification, the BoW method records the number of occurrences of each bag that is created for each instance type or word disregarding the order of the words or the grammar. And in visual scene classification it is based on clusters of local descriptors which are taken from the images disregarding the order of the clusters. The key idea is to generate a histogram for the words in the documents or the features in the images to represent the specified document or image. The BoW method is computationally and even conceptually is simpler than many other classification methods. For that reason, BoW based systems could record new and higher performance scores on common used benchmarks of text and image classification algorithms. This paper presents an overview of BoW, importance of BoW, how does it work, applications and challenges of using BoW. This study is useful in terms of introducing the BoW method to the new researchers and providing a good background with associated related works to the researchers that are working on the model.

This research introduces an approach to diagnose diseases from medical check-up test reports. The proposed approach is produced from Optical Character Recognition (OCR) technology to convert the hard copy test reports into editable textual data, Bag of Words (BoW) model as feature selection algorithm, Naive Bayes as classification algorithm, and AdaBoost technique to enhance the performance of the Naive Bayes classifier. The performance of the proposed approach is very good in terms of validity and can be used in diagnosing of diseases from medical check-up test reports. The proposed approach is trained on dedicated trained partitions of multiple medical datasets, and then tested on the testing sets partitioned from the original datasets. The proposed algorithm is compared with the Support Vector Machine (SVM), Naive Bayes (NB), Decision Table (DT), and k-Nearest Neighbors (k-NN) classifiers, in which all the algorithms are tested on the same datasets. The proposed algorithm showed higher accuracy than the other four classifiers. So, the proposed approach which is the combination of BoW with AdaBoost technique is used to predict the name of the diseases from the medical check-up test reports. After that, an image as an example of the disease will be presented as well with the name of the disease to the physician and the patient. The image presentation is very important for the patients, because they may not familiar with the medical terms and disease names. Finally, the proposed approach can be used in the medical area because of its good performance and showing validated results after it is tested.

Content-based image retrieval (CBIR) is an application of machine learning used to retrieve images by similarity on the basis of features. Our objective was to develop a CBIR system that could identify images containing the same polyp ('polyp fingerprint').A machine learning technique called Bag of Words was used to describe each endoscopic image containing a polyp in a unique way. The system was tested with 243 white light images belonging to 99 different polyps (for each polyp there were at least two images representing it in two different temporal moments). Images were acquired in routine colonoscopies at Hospital Clínic using high-definition Olympus endoscopes. The method provided for each image the closest match within the dataset.The system matched another image of the same polyp in 221/243 cases (91%). No differences were observed in the number of correct matches according to Paris classification (protruded: 90.7% vs. non-protruded: 91.3%) and size ( 10 mm: 91.6% vs. 10 mm: 90%).A CBIR system can match accurately two images containing the same polyp, which could be a helpful aid for polyp image recognition.

Knowledge graphs such as DBPedia and Freebase contain sparse linkage connectivity, which poses severe challenge to link prediction between entities. In addressing this sparsity problem, our studies indicate that one needs to leverage model with low complexity to avoid overfitting the weak structural information in the graphs, requiring the simple models which can efficiently encode the entities and their description information and then effectively decode their relationships. In this paper, we present a simple and efficient model that can attain these two goals. Specifically, we use a bag-of-words model, where relevant words are aggregated using average pooling or a basic Graph Convolutional Network to encode entities into distributed embeddings. A factorization machine is then used to score the relationships between those embeddings to generate linkage predictions. Empirical studies on two real datasets confirms the efficiency of our proposed model and shows superior predictive performance over state-of-the-art approaches., The Web Conference 2019, May 13-17, 2019, San Francisco, California, United States

View synthesis is the most important technique in multi-view and free-viewpoint videos. The whole view synthesis includes the acquisition and processing of texture and depth images, and the virtual view rendering stage. Existing quality metrics for view synthesis have limited ability for the whole synthesis process for the following reasons. First, they are dedicated to a single stage of view synthesis, overlooking the commonality of all the possible distortions introduced in the whole process. Moreover, they only extract low-level features for quality assessment, ignoring the perceptual degradation caused by the mid-level contours that are destructed by heavy distortions in texture/depth images and the imperfect view rendering, which represent the spatial distribution/connection of adjacent contour pixels. Inspired by the above facts, this paper presents a quality metric for view synthesis using both Low-level and Mid-level Structural representation (LMS), aiming to accurately evaluate the distortions in the whole view synthesis process. Specifically, the scale space is first constructed to mimic the hierarchical property of the human visual system. Then, the statistics of gradient orientation is integrated with the statistics of gradient intensity for the low-level structural representation, which is motivated by the importance of the orientation selectivity mechanism to visual perception. Further, the mid-level structure is represented using bag of words for contour description based on the sparse coding of the primary visual cortex. Then the distances of both the low-level and mid-level features between the synthesized and reference images are calculated. Finally, two distances are integrated to generate the whole quality score. Extensive experiments on two public view synthesis databases demonstrate the superiority of the proposed method to the state-of-the-arts in evaluating the quality of the whole view synthesis.

The security analysis of sensitive issues and medical diagnosis are immensely focus to determine exact location of event happening regions. In this paper we propose a model of clustering and pooling techniques to local features using Bag-of-Words (BoW) descriptor in SVM framework for event detection in video sequences. The proposed model extracts local features from six categories of Columbia Consumer Video (CCV) event detection benchmark. We developed the clusters of these features using KD-search tree and Lloyds algorithm. The clusters of features is pooled to vectors by using bag of words model. Introducing the inferring temporal instance labelling, the model performed fast for event detection. The significant performance of the research problem can thrilled out the social media by retrieving the best possible content. The proposed model can efficiently perform the experiments of event detection related to big data problem in visual media. Furthermore, the proposed approach in the model is invariant to rotation, translation and scale changes in the video sequence and robust to the illumination and viewpoints.

Crime records hold critical information about crime situations including the offender actions, stolen objects description, characteristics of victims, the location of the crime situation, individuals involved and more. To consume crime data, police forces and other security analysts use risk management systems which process and organize data, and summarize it into relevant and useful information on criminal situations. This type of system depends on promoting Situation Awareness to stimulate the users understanding of the crime situations and consequently the decision-making assertiveness. In this work the goal is to contribute with the typification of crime situations by machine learning driven techniques, applied in conjunction with pre-processing and transformation. Results showed that the use of pre-processing techniques improved data quality and algorithms precision. In addition, the transformation technique with the best results found was Bag of Words Binarization. Finally, the Logistic Regression algorithm presented the best results for mining the crime data.

This paper discusses about the various approaches that has been taken to reduce downtime in Coal bed methane (CBM) wells by predicting the string integrity failure in advance. Various approaches such as the Principal component analysis (PCA) based T-statistics approach and the Bag of Features approach have been taken to find a solution. These approaches fall under the classical classification approach of supervised learning. The power and usefulness of these approaches are fuelled and limited by the number and richness of annotations and sensors at the Well site. These approaches used on the CBM wells gave very encouraging results, thereby proving the helpfulness of this approach in enhancing the well efficiency and decreasing the well downtime by planning based on the failure prediction of the model.

When natural disasters strike, annotated images and texts flood the Internet, and rescue teams become overwhelmed to prioritize often scarce resources, while relying heavily on human input. In this paper, a novel multi-modal approach is proposed to automate crisis data analysis using machine learning. Our multi-modal two-stage framework relies on computationally inexpensive visual and semantic features to analyze Twitter data. Level I classification consists of training classifiers separately on semantic descriptors and combinations of visual features. These classifiers' decisions are aggregated to form a new feature vector to train the second set of classifiers in Level II classification. A home-grown dataset is gathered from Twitter to train the classifiers. Low-level visual features achieved an accuracy of 91.10% which increased to 92.43% when semantic attributes were incorporated. Applying such data science techniques on social media seems to motivate an updated folk statement "an ANNOTATED image is worth a thousand words".

In this paper, we describe our submission to SemEval-2019 Task 4 on Hyperpartisan News Detection. Our system relies on a variety of engineered features originally used to detect propaganda. This is based on the assumption that biased messages are propagandistic in the sense that they promote a particular political cause or viewpoint. We trained a logistic regression model with features ranging from simple bag-of-words to vocabulary richness and text readability features. Our system achieved 72.9% accuracy on the test data that is annotated manually and 60.8% on the test data that is annotated with distant supervision. Additional experiments showed that significant performance improvements can be achieved with better feature pre-processing., Hyperpartisanship, propaganda, news media, fake news, SemEval-2018

The proposed system is going to deal with a very challenging task of automatically generating presentation slides for academic papers. The wide accessibility of web archives in electronic structures requires a programmed method to mark the records with a predefined set of subjects, what is known as customized Text Categorization (TC). Over the previous decades, it has been seen a substantial number of cutting edge machine learning calculations to address this testing errand. The produced introduction slides can be used as drafts to enable the moderators to set up their formal slides quickly. Documents are usually represented by the "bag-of-words": namely, each word or phrase occurs in documents once or more times is considered as a feature. It initially utilizes the relapse strategy to take in the significance scores of the sentences in a scholastic paper, and afterward a compelling calculation is created for multi-name grouping with using those information that are important to the objectives.The key is the development of a coefficient-based mapping among preparing and test cases, where the mapping relationship abuses the connections among the examples, instead of the unequivocal connection between the factors and the class marks of information and fabricates the staggered classifier on the adjusted low-dimensional data depictions in the meantime. It at first uses the backslide system to take in the importance scores of the sentences in an educational paper, and after that experiences the Latent Dirichlet Allocation (LDA) methodology to make especially sorted out slides by picking and modifying key articulations and sentences to a point for the slide. We set up a sentence scoring model in light of gullible Bayes classifier and use the LDA strategy to modify and expel key articulations and sentences for delivering the slides. Exploratory results exhibit that our technique can deliver very much wanted slides over regular procedures.

Content Based Image Retrieval (CBIR) is a challenging ground to retrieve the images based on feature description and extraction. Salient image features are the potential candidates to search and index the images based on primitive features. This paper presents a novel way to extract and describe the image features based on their object and color signatures. Gaussian of variance is used along with image convolution and the resultant features are approximated using matrix hashing. The high range frequency feature information is then separated to carry out the spatial indexes. High resolution images from versatile image semantic groups are resulted in massive image signatures which are computation effective. Principal component analysis is performed on these feature vectors to obtained reduced statistical information. These grey level information carriers are then combined with RGB channel based spatially arranged features to possess to color and object information. Bags-of-Word (BoW) architecture is applied to these information carriers to index and retrieve the images. Competitive results show that the proposed method outperforms in many categories of benchmarks Caltech-101 and Corel-1000 datasets. The proposed method shows outstanding results in complex and occluded image categories.

Antibody development is crucial for immunohistochemistry (IHC) applications. To improve the efficiency of primary antibody screening processes, we developed a computer aided detection scheme to automatically identify the non-negative tissue slides which indicate reactive antibodies. A dataset with 564 digital IHC whole slide images were used for algorithm training and testing, each of which was labeled by pathologist as a negative (i.e., no staining) or non-negative (i.e., pure background or partial staining) slide. To avoid unnecessary computations, color deconvolution was first applied to low resolution whole slide images and histogram based image features were extracted from each unmixed single stain image. Then, different classifiers were built using the low resolution image features computed from the training dataset through ten-fold cross validation. The trained model was tested over the testing dataset. Results indicated that linear supported vector machine (LSVM) method yielded the highest area under ROC curve. To further improve the accuracy, our scheme utilized the LSVM classifier score to identify the slides for which additional analysis was needed. The additional analysis was performed through dividing the original whole slide image into non-overlapping tiles and extracting high resolution image features from each tile. The tile-based features are then used to form a bag-of-words (BoW) representation of the corresponding whole slide image, based on which a second classifier was built to perform the predictions. The results showed that the proposed scheme can effectively perform negative versus non-negative classification with high accuracy and thus reduce pathologists’ manual reviewing time for antibody screening.

Development of information technology makes the production of data increase dramatically. We can get lots of data from the internet, including data reviews about a product or service. The more data obtained, the system is needed to process it. Sentiment analysis is a text processing of Natural Language Processing (NLP) that can help someone to see the quality of service offered, including hotel services. This paper uses hotel review data to carry out sentiment analysis obtained from the Traveloka website. The data classified using the Long Short-Term Memory (LSTM) algorithm. To get better results, the authors use word embedding to convert words into vectors. This study aims to compare the performance of several word embedding, while word embedding compared is word2vec Continuous Bag of Words CBOW, word2vec skip-gram, doc2vec, and glove. From the experiment conducted, the results show that the glove method has the highest accuracy of 95.52% while the word2vec skip-gram model has the lowest accuracy of 91.81%, so it concluded that the glove method is the best word embedding method for hotel review data.

People use social media for entertainment, fetching information, news, business, communication and many more. Few of such social media applications are Facebook, Twitter, WhatsApp, Snapchat and so on. Twitter is one among the micro blogging websites. We are using Twitter mainly because it has gained a lot of media attention. The text written is referred to as tweets, where a common man can tweet or can write their hearts out. We would be fetching the direct responses from the public and hence the data is more real-time. First step is to fetch the tweets on a particular scheme using python language code followed by the cleaning process then comes the creation of bag of words. Later these bags of words are given as an input to the algorithms. Finally, after training the algorithms, we will be getting the sentiment of the public on that scheme.

Authorship Attribution (AA), is a process to identify an author based on input text data given to the system based on its characteristics is a problem with a long history. In this project, we study the problem of authorship attribution for forensic purposes and present machine learning techniques and stylometric features of the author tweets. For this purpose micro-blogging site Twitter is taken for experimentation purpose. On this site people share their ideas, likes, dislikes, interest, opinion, thoughts in the form of short messages called tweets. More than thousand tweets are posted every second and the possibility of sensitive, illicit text sharing cannot be ignored. This system downloads live twitter tweets, and takes text file as the input. The text file contains tweets of random author. Our system finds that that tweet downloaded belongs to which author. For classification of the author some important features are used. Important features include calculation of smiley, calculation of stop words, calculation of punctuations, and calculation of similarity words. Basically this system is divided into two stage process, where in the first stage, stylometric information is extracted from the collected dataset and in the second stage classification algorithm is trained to predict authors of unseen text. The effort is to find out which combination of features help in accurate prediction of the author thus maximizing the accuracy of predictions with optimum amount of data.

Semantic gap optimization is the major objective of Content Based Image Retrieval(CBIR) system. It is challenging to derive possible query objectives in case of an image having multiple objects or complex background. The query requirements are defined as the Bag of Words(BoW) and used for retrieval of the similar images. The saliency based segmentation is popularly used to eliminate the background objects that works well when the image is focused on the foreground objects. In case saliency-map fail to extract the foreground object, the feature vectors are equally biased by the background scene and that imposter the actual query objectiveThis paper proposes a two phase framework CBIR system. In the first phase we compute the Local Binary Pattern and the RGB Colour information(LBPC) features of all the images in the large image database. Then the pertinent images are clustered and the image on the cluster centre is assumed as ideal indexed image for the same class. In the second phase, each input image is divided into nine blocks having common central sub-block. The LBPC feature is extracted for all the blocks and only the feature of the block that best matches with the indexed image is stored in the database. The selection of the best block image feature minimizes the effect of the background scene. This approach helps select the potential saliency block of the image with respect to the indexed image. The result section shows the performance of the proposed framework while BoW is defined by the cluster indexed image.

Question classification helps to generate more accurate answers in question answering system. For an efficient question classifier, one of the most important tasks is to fully mine useful features. Aiming at solving the problem of lacking of rich syntax and semantic features in Chinese question classification, an operator called MBWB (multilayer bag-of-words binding) is proposed to extract potential features by binding part-of-speech, word sense, named entity and other basic features to bag-of-words, respectively. Through performing MBWB operator on two kinds of bag-of-words, i.e. A_BOW and T_BOW, the corresponding A_MBWB and T_MBWB features are generated automatically. The MBWB operator can explore potential information contained in basic features, and enrich syntactic and semantic representation of questions. Experimental results on the Chinese question set show that the classification accuracy gets significantly improved when combining two kinds of MBWB features with basic features.

Fake data detection is the most important problem to be addressed in the recent years, there is lot of research going on in this field. Because of its serious impacts on the readers. researchers, government and private agencies working together to solve the issue. This paper represents a hybrid approach for fake data detection using the multinomial voting algorithm. This algorithm was tested with multiple fake news dataset which resulted in an accuracy score of 94 percent which is a benchmark in the machine learning field where the other algorithms are at a range of 82 to 88 percent. The list of algorithms that have been used here is as follows Naive Bayes, Random Forest, Decision Tree, Support Vector Machine, K Nearest Neighbors. All these algorithms use training data as the bag of words model which was created using Count Vectorizer. Experimental data has collected from the Kaggle data world. Python is used as a language to verify and validate the results. Tableau is used as a visualization tool. Implementation is carried out using default algorithm values.

As the recent advancement in communication technologies revolutionizes the world, Email is emerging as a wide-known communication paradigm in various business processes. Email is an effective, brisk and minimal effort correspondence approach. Email Spam is non-asked for information sent to the E-letter drops. Spam could be an enormous downside each for clients and for ISPs. As per examination these days client gets a considerable measure of spam messages then non-spam messages. In some cases spam messages may damage the reputation of a particular business process. It can be observed that in most of the popular mailing services the spam filters are being biased for the profit from the ads, i.e. they are allowing some exception for some companies that pay for advertising. This is not an ethical practice, but it is very profitable for their future business processes. Our aim is to build a spam detector using machine learning in python with the packages NLTK, Matplotlib, Word cloud, Math, pandas, NumPy. With this proposed model, we can state a specified message as spam or non-spam. It can be implemented by using Bayes’ Theorem, a simple yet powerful theorem.

Handwriter identification aims to simplify the task of forensic experts by providing them with semi-automated tools in order to enable them to narrow down the search to determine the final identification of an unknown handwritten sample. An identification algorithm aims to produce a list of predicted writers of the unknown handwritten sample ranked in terms of confidence measure metrics for use by the forensic expert will make the final decision. Most existing handwriter identification systems use either statistical or model-based approaches. To further improve the performances this paper proposes to deploy a combination of both approaches using Oriented Basic Image features and the concept of graphemes codebook. To reduce the resulting high dimensionality of the feature vector a Kernel Principal Component Analysis has been used. To gauge the effectiveness of the proposed method a performance analysis, using IAM dataset for English handwriting and ICFHR 2012 dataset for Arabic handwriting, has been carried out. The results obtained achieved an accuracy of 96% thus demonstrating its superiority when compared against similar techniques. This work is supported by the Qatar National Research Fund through National Priority Research Program (NPRP) No 7-442-1-082 . The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of the Qatar National Research Fund or Qatar University. Scopus

Text mining is one of the main and typical tasks of machine learning (ML). Authorship identification (AI) is a standard research subject in text mining and natural language processing (NLP) that has undergone a remarkable evolution these last years. We need to identify/determine the actual author of anonymous texts given on the basis of a set of writing samples. Standard text classification often focuses on many handcrafted features such as dictionaries, knowledge bases, and different stylometric characteristics, which often leads to remarkable dimensionality. Unlike traditional approaches, this paper suggests an authorship identification approach based on automatic feature engineering using word2vec word embeddings, taking into account each author's writing style. This system includes two learning phases, the first stage aims to generate the semantic representation of each author by using word2vec to learn and extract the most relevant characteristics of the raw document. The second stage is to apply the multilayer perceptron (MLP) classifier to fix the classification rules using the backpropagation learning algorithm. Experiments show that MLP classifier with word2vec model earns an accuracy of 95.83% for an English corpus, suggesting that the word2vec word embedding model can evidently enhance the identification accuracy compared to other classical models such as n-gram frequencies and bag of words.

Social media became prevalent in our day to day life. People express their views, feelings, emotion via comments. Those comments entered by the user is analyzed. The Naive Bayes Algorithm is used to detect the offensive and sarcastic comment, thus preventing them from being uploaded in social media.

We investigate the use of Twitter data to deliver signals for syndromic surveillance in order to assess its ability to augment existing syndromic surveillance efforts and give a better understanding of symptomatic people who do not seek health care advice directly. We focus on a specific syndrome - asthma/difficulty breathing. We outline data collection using the Twitter streaming API as well as analysis and pre-processing of the collected data. Even with keyword-based data collection, many of the tweets collected are not be relevant because they represent chatter, or talk of awareness instead of suffering a particular condition. In light of this, we set out to identify relevant tweets to collect a strong and reliable signal. For this, we investigate text classification techniques, and in particular we focus on semi-supervised classification techniques since they enable us to use more of the Twitter data collected without needing to label it all. In this paper, propose a semi-supervised approach to symptomatic tweet classification and relevance filtering. We also propose the use of emojis and other special features capturing the tweet’s tone to improve the classification performance. Our results show that negative emojis and those that denote laughter provide the best classification performance in conjunction with a simple bag of words approach. We obtain good performance on classifying symptomatic tweets with both supervised and semi-supervised algorithms and found that the proposed semi-supervised algorithms preserve more of the relevant tweets and may be advantegeous in the context of a weak signal. Finally, we found some correlation (r = 0.414, p = 0.0004) between the Twitter signal generated with the semi-supervised system and data from consultations for related health conditions.

After document and image retrieval, shape retrieval is seen as the next step in the search process. Numerous techniques have been proposed to retrieve shapes meta-data considering shapes similarity but these techniques normally suffer from non-interactive running times. A widely used approach, which has its origins in document retrieval, and is successfully been applied for image retrieval, is based on "bag-of-words". Methods using this approach have the potential to support fast, real time retrieval of shapes over the large databases. This research examines these approaches and summarizes the techniques used to convert shapes to their "bag of words" representations. This research also explores the mechanism these techniques adopts to facilitate and enhance interactive search times.

Data mining techniques have seen tremendous increase in their usage in the past few years. Patent mining is one of the domains that utilize data mining techniques to a great extent. Patent mining consists of various tasks such as retrieval of patent, classification, patent valuation, patent visualization and detecting infringements. Among these, patent classification is an important task. It deals with the classification of patents into various categories. A common bottleneck in this task has been related to the automated classification of patents with better accuracy. The rapid increase in the number of patents being filed every year and the increasing complexity of the patent documents demand for advanced and revolutionized tools/machines to assist in performing patent classification in automated manner. Usually, the patents are examined thoroughly by patent analysts from various domains, who possess respective expertise and are well aware of the domain jargons. The main objective of such systems is to get rid of the time-consuming, laborious manual process and to provide patent analysts a better way for classifying patent documents. Also it helps in better management, maintenance and convenient searching of patent documents. Here, two prominent classification algorithms—Naive Bayes and support vector machines (SVM)—are explored and implemented. Additionally, some pre-processing steps such as stop word removal, stemming, and lemmatizing are also done to obtain better accuracy. TF-IDF feature is also incorporated to obtain precise results.

Online notepad services allow users to upload and share free text anonymously. Reviewing Pastebin, one of the most popular online notepad services websites, it is possible to find textual content that could be related to illegal activities, such as leaks of personal information or hyperlinks to multimedia files containing child sexual abuse images or videos. An automatic approach to monitor and to detect these activities in such an active and a dynamic environment could be useful for Law Enforcement Agencies to fight against cybercrime. In this work, we present Pastes Content Classification 17K (PasteCC_17K), a dataset of 17640 textual samples crawled from Pastebin, which are classified in 15 categories, being 6 of them suspicious to be related to illegal ones. We used PasteCC_17K to evaluated two well-known text representation techniques, ensembled with three different supervised approaches to classify the pastes of the Pastebin website. We found that the best performance is achieved ensembling TF-IDF encoding with Logistic Regression obtaining an accuracy of \(98.63\%\). The proposed model could assist the authorities in the detection of suspicious content shared in Pastebin.

The consumer reviews serve as feedback for busi-nesses in terms of performance, product quality, and consumer service. In this research, we predict consumer opinion based on mobile phone reviews, in addition to providing an analysis of the most important factors behind reviews being classified as either positive, negative, or neutral. This insight could help companies improve their products as well as helping potential buyers to make the right decision. The research presented in this paper was carried out as follows: the data was pre-processed, before being converted from text to vector representation using a range of feature extraction techniques such as bag-of-words, TF-IDF, Glove, and word2vec. We study the performance of different machine learning algorithms, such as logistic regression, stochastic gradient descent, naive Bayes and convolutional neural networks. In addition, we evaluate our models using accuracy, F1-score, precision, recall and log loss function. Moreover, we apply Lime technique to provide analytical reasons for the reviews being classified as either positive, negative or neutral. Our experiments revealed that convolutional neural network with word2vec as a feature extraction technique provides the best results for both the unbalanced and balanced versions of the dataset.

In this paper, we propose a human action recognition method using HOIRM (histogram of oriented interest region motion) feature fusion and a BOW (bag of words) model based on AP (affinity propagation) clustering. First, a HOIRM feature extraction method based on spatiotemporal interest points ROI is proposed. HOIRM can be regarded as a middle-level feature between local and global features. Then, HOIRM is fused with 3D HOG and 3D HOF local features using a cumulative histogram. The method further improves the robustness of local features to camera view angle and distance variations in complex scenes, which in turn improves the correct rate of action recognition. Finally, a BOW model based on AP clustering is proposed and applied to action classification. It obtains the appropriate visual dictionary capacity and achieves better clustering effect for the joint description of a variety of features. The experimental results demonstrate that by using the fused features with the proposed BOW model, the average recognition rate is 95.75% in the KTH database, and 88.25% in the UCF database, which are both higher than those by using only 3D HOG+3D HOF or HOIRM features. Moreover, the average recognition rate achieved by the proposed method in the two databases is higher than that obtained by other methods.

2018 International Conference on Artificial Intelligence and Data Processing, IDAP 2018 --28 September 2018 through 30 September 2018 -- -- Text categorization is a text mining process and it aims to discover relevant information and relationship in a huge amounts of text data. Feature extraction is an important preprocessing step of text categorization, as extracted features are used to represent texts. Several methods, feature models, and algorithms are needed to extract useful features from textual contents. One of the these methods is frequent itemset mining which is a basic data mining technique that employed to find interesting patterns in data. As the frequent itemsets (termsets) reflect strong associations between items they provide more underlying contextual semantic than an individual word. Therefore, it is used in text mining domain for different purposes (e.g., frequent itemset based text clustering). In this paper, we employ termsets for text representation, and use both binary and cardinality based approaches for termset weighting. Unlike existing studies, we use character level n-grams to represent items in a transaction in addition to the traditional bag of words model. Through our experimental results on Mod-Apte split of the Reuters-21578 dataset, we determined that performing document-transaction conversion at level of character n-grams improves the performance of the Support Vector Machine classifier. © 2018 IEEE.

We propose a new method for anomaly detection of human actions. Our method works directly on human pose graphs that can be computed from an input video sequence. This makes the analysis independent of nuisance parameters such as viewpoint or illumination. We map these graphs to a latent space and cluster them. Each action is then represented by its soft-assignment to each of the clusters. This gives a kind of "bag of words" representation to the data, where every action is represented by its similarity to a group of base action-words. Then, we use a Dirichlet process based mixture, that is useful for handling proportional data such as our soft-assignment vectors, to determine if an action is normal or not. We evaluate our method on two types of data sets. The first is a fine-grained anomaly detection data set (e.g. ShanghaiTech) where we wish to detect unusual variations of some action. The second is a coarse-grained anomaly detection data set (e.g., a Kinetics-based data set) where few actions are considered normal, and every other action should be considered abnormal. Extensive experiments on the benchmarks show that our method performs considerably better than other state of the art methods., Comment: Code is available at https://github.com/amirmk89/gepc. CVPR 2020

We propose a fast Bag-of-Words (BoW) method for image classification, inspired by the mechanism that arrangement of neurons in visual cortex can preserve the topology of mapping from inputs, and the fact that human brain can retrieve information almost instantly. We propose algorithms for accelerating both Self-Organizing Map (SOM) training and BoW coding. First, we modify the traditional SOM based on the matrix factorization form of K-means. Utilizing the topology-preserving property of dictionary learned by SOM, the coding process of BoW can be accelerated by fast search of k-nearest neighbor codewords in the grid of SOM dictionary. We evaluate the proposed method in different coding scenarios for image classification task on MNIST and CIFAR-10 datasets. The results show that the proposed method accelerates BoW classification greatly with little loss of classification accuracy.

This paper presents a named entity recognition method which finds predetermined entities in an unstructured text. The method uses word similarities based on typical word transformations (lemmatization and stemming), word embeddings and character level based similarity to map those entities onto words in the text. The approach is language independent, though language-dependent components are used for lemmatization, stemming and word embedding, and works on any given set of entities. Special attention is given to the entities which are represented in a hierarchical form with the hypernymy-hyponymy relation. The proposed method has the following advantages: it finds the normalized form of the recognized entity name; it is easy to adjust to a new domain; it respects the hierarchical organization of entities; and due to the modular approach can be constantly improved just by updating components for lemmatization, stemming or word embedding. The proposed entity recognition method was tested on a test set of tourist queries and hierarchical entities collected from Slovenia.info tourist portal.

Parkinson’s disease (PD) is a neuro-degenerative disorder that produces motor and non-motor impairments. Non-motor impairments include communication and mood disorders. Most of the studies in the literature have been focused on the analysis of motor symptoms. However, non-motor signs are also present in most of the cases. This paper addresses the study of language production as a potential tool to diagnose and monitor the neurological state of PD patients. The study proposes the use of natural language processing methods to extract features from transcriptions obtained from spontaneous speech recordings to discriminate between healthy control people and PD patients. The analysis considered classical features such as Bag of Words and Term Frequency-Inverse Document Frequency, along with methods based on word-embeddings. Accuracies of up to 72% are obtained when discriminating between PD patients and healthy subjects, which confirms that there is information embedded in the language production that can be used for the assessment of the disease.

Recommendation systems are achieving great success in e-Commerce applications, during a live interaction with a customer; recommendation system may apply different techniques to solve the problem of making a correct and relevant product recommendation. The main objective of this research is to perform product recommendation using textual similarity based Learning model. In this research data acquired through Amazon product advertising API after Data cleaning and text preprocessing the content based product recommendation have been performed using Bag of Words(BOW) and Term Frequency-Inverse Document Frequency(TF-IDF) based text vectorization techniques. Textual Description of the product converted into n-dimensional vector, and later the Euclidean similarity can be measured between the ndimensional vector of the queried product and other products. Text-based product similarity through text vectorization technique is very useful in performing content-based product recommendation and recommending the similar item to the user it can be used in various E-Commerce applications since these applications are heavily populated with the textual description of the product. Bag of words and TF-IDF generates an n-dimensional vector of a different textual description of the product which in turn leads to a better recommendation of the product. Experimental results and analysis section clearly describes how the proposed model for text-based product find the similarity of products with queried product and display as output the best-recommended product.

We investigate two different approaches to text classification, by categorising enquiries submitted to the House of Commons Library from elected Members of the UK Parliament. One is an unsupervised approach, i.e. topic modelling, and the other is a supervised approach based on weakly labelled data, i.e. distant supervision. Models were trained on two types of feature sets: one based only on bag of words, and the other combining bag of words with structured metadata attached to enquiries. Our results show that topic modelling obtains superior performance on this task, and that the incorporation of structured metadata as learning features contributes insignificantly to improved model performance.

Background: Text classification is a very important task in information retrieval. Its objective is to classify new text documents in a set of predefined classes, using different supervised algorithms. Objectives: We focus on the text classification for Albanian news articles using two approaches. Methods/Approach: In the first approach, the words in a collection are considered as independent components, allocating to each of them a conforming vector in the vector’s space. Here we utilized nine classifiers from the scikit-learn package, training the classifiers with part of news articles (80%) and testing the accuracy with the remaining part of these articles. In the second approach, the text classification treats words based on their semantic and syntactic word similarities, supposing a word is formed by n-grams of characters. In this case, we have used the fastText, a hierarchical classifier, that considers local word order, as well as sub-word information. We have measured the accuracy for each classifier separately. We have also analyzed the training and testing time. Results: Our results show that the bag of words model does better than fastText when testing the classification process for not a large dataset of text. FastText shows better performance when classifying multi-label text. Conclusions: News articles can serve to create a benchmark for testing classification algorithms of Albanian texts. The best results are achieved with a bag of words model, with an accuracy of 94%.

Many architectures for multi-task learning (MTL) have been proposed to take advantage of transfer among tasks, often involving complex models and training procedures. In this paper, we ask if the sentence-level representations learned in previous approaches provide significant benefit beyond that provided by simply improving word-based representations. To investigate this question, we consider three techniques that ignore sequence information: a syntactically-oblivious pooling encoder, pre-trained non-contextual word embeddings, and unigram generative regularization. Compared to a state-of-the-art MTL approach to textual inference, the simple techniques we use yield similar performance on a universe of task combinations while reducing training time and model size.

Social media (such as Twitter) helps users express their opinions easily through writings. With the increase in the use of social media, the amount of textual data coming from direct users has grown enormously. These data can help to extract meanings of the specific situations using text mining. For example, in moments of disaster, social media can be used to get important information about the current situation. Understanding the emotions of the users can help humanitarian aid to be allocated more effectively. Sentiment analysis is a method used to get emotion rank from the textual data. It helps get the numerical scale of the emotion from the unstructured data. This study aims to create sentiment analysis, also known as opinion mining, from the given data by applying both machine learning and deep learning methods for disaster moments. We intend to see which technique is more successful in terms of sentiment analysis through comparison of the methods. We created a Turkish sentiment analysis about the coup attempt that happened on July 15, 2016. In this work, the coup attempt is considered a disaster as it consisted of violence and happened suddenly. The dataset used is comprised of unstructured sentences and labels with sentiment points that are either positive or negative. In this work, we pre-process the dataset and apply Turkish Natural Language Processing (NLP). Three text-features (Bag of Words, Tf-Idf, Word 2Vec) are used to vectorize and prepare the dataset for the application of the models. Then, we apply machine learning and deep learning methods and compare them while looking at the results. Examining the results and comparison, we aim to see whether social media can help disaster management. In this comparison, we observed that deep learning has higher accuracy. Our findings indicated that social media can be used to understand sudden situations like disasters and may help to decide how to act.

Targeted email attacks are one of main threats for organizations of all sizes and across every field. In targeted email attacks, malicious VBA (Visual Basic for Applications) macros are often contained in the attachment files to exploit the target computers. These malicious VBA macros are obfuscated in several ways to evade detection. Hence, pattern-based detection has a limitation in detecting these new malicious VBA macros. To detect new malicious VBA macros, some methods with machine learning techniques have been proposed. A method extracts words from the source code, and constructs a language model to represent VBA macros for machine learning techniques. This method, however, constructs a language model from all the extracted words. Therefore, this model might contain unnecessary words to classify. To construct an efficient language model, we focus on LSI (Latent Semantic Indexing). LSI is one of the foundational techniques in topic modeling, and calculates similarity of documents. Our method uses LSI to construct an efficient language model, which produces more accuracy and efficiency. To the best of our knowledge, our method is the first method to detect new malicious VBA macros with LSI. Our method extracts words from the source code and converts into feature vectors with some Natural Language Processing techniques. Our method trains a classifier with benign and malicious VBA macros and detects new malicious VBA macros. Several thousands of samples for evaluation are obtained from Virus Total. The experimental result shows that our method can detect new malicious VBA macros more accurately and efficiently. The best F-measure achieves 0.95.

User reviews submitted to app marketplaces contain information that falls into different categories, e.g., feature evaluation, feature request, and bug report. This information is valuable for developers to improve the quality of mobile applications. However, due to the large volume of reviews received every day, manual classification of user reviews into these categories is not feasible. Therefore, developing automatic classification methods using machine learning approaches is desirable. In this study, we address the problem of automatic classification of app review sentences (as opposed to full reviews) into different categories. We compare the simplest textual machine learning classifier using only lexical features – the so-called Bag-of-Words (BoW) approach – with more complex models used in previous work adopting rich linguistic features. We find that the performance of the simple BoW model is very competitive and has the advantage of not requiring any external linguistic tools to extract the features. Moreover, we experiment with deep learning based Convolutional Neural Network (CNN) models that have recently achieved state-of-the-art results in many classification tasks. We find that, on average, the CNN models do not perform significantly better than the simple BoW model. Finally, the manual analysis of misclassification errors and data annotations suggests that classifying review sentences in isolation does not always contain enough information to make a correct prediction. Thus, we suggest that adopting neural models to incorporate additional contextual knowledge might improve the classification performance.

Nowadays gender classification is a very challenging task in a real-time application based on face recognition. The demand for real-time application based on gender classification will increase in the future. Bag of Words (Bow), Scale Invariant Fourier Transform (SIFT) and K-means clustering are used for feature extractors and classification. This state of art methodology gives more efficient result on different standard datasets. This research proposed a new algorithm for automatic live Gender Recognition (GR) using Support Vector Machine (SVM) is used for classification. The implementation of result work tested on FEI, Live Images and SCIEN database for GR. The detection rate has reached up to 98% in FEI dataset, 94% in Live/Own dataset and 91% SCIEN dataset respectively. This proposed state of art methodology is compared with the previous techniques and achieved better results which will help in the development of real-time identification systems.

We propose a novel topic discovery algorithm for unlabeled images based on the bag-of-words (BoW) framework. We first extract a dictionary of visual words and subsequently for each image compute a visual word occurrence histogram. We view these histograms as rows of a large matrix from which we extract sparse principal components (PCs). Each PC identifies a sparse combination of visual words which co-occur frequently in some images but seldom appear in others. Each sparse PC corresponds to a topic, and images whose interference with the PC is high belong to that topic, revealing the common parts possessed by the images. We propose to solve the associated sparse PCA problems using an Alternating Maximization (AM) method, which we modify for the purpose of efficiently extracting multiple PCs in a deflation scheme. Our approach attacks the maximization problem in SPCA directly and is scalable to high-dimensional data. Experiments on automatic topic discovery and category prediction demonstrate encouraging performance of our approach. Our SPCA solver is publicly available.

We here present an operational deep learning pipeline for classifying life events from individual tweets, using job loss as a use case and Twitter data collected between 2010 and 2013 (historic sample from the public stream). The pipeline includes identification of keywords through snowball sampling, multiple rater manual annotation, supervised deep learning, text processing (word embedding, bag of words) and architecture selection (convolutional, shallow-and-wide convolutional, and long-short-term memory) with parameter optimization, external validation and feedback learning. After model optimization, a shallow-and-wide network with a pre-trained 200-dimensional word2vec achieved a precision of 78% (over an average single keyword precision of 50%) and an area under receiver operating characteristic of 86%. Precision and recall also increased by 5% using bag of words. When tested on tweets with ambiguous annotations (i.e. tweets that were hard for human annotators to classify), the network achieved 65% precision. Finally, on a random set of tweets that did not contain any of the snowballed keywords, 30% were classified as job loss events; this putatively false positive set can be used to reinforce the learner’s training. In conclusion, the pipeline streamlines both the manual and automated process, providing feedback reinforcement (snowballing and external tweets), and shows good performance on classifying individual tweets on the use case, potentially saving human resources needed to collate such data for research studies.

Accurate semantic representation models are essential in text mining applications. For a successful application of the text mining process, the text representation adopted must keep the interesting patterns to be discovered. Although competitive results for automatic text classification may be achieved with traditional bag of words, such representation model cannot provide satisfactory classification performances on hard settings where richer text representations are required. In this paper, we present an approach to represent document collections based on embedded representations of words and word senses. We bring together the power of word sense disambiguation and the semantic richness of word- and word-sense embedded vectors to construct embedded representations of document collections. Our approach results in semantically enhanced and low-dimensional representations. We overcome the lack of interpretability of embedded vectors, which is a drawback of this kind of representation, with the use of word sense embedded vectors. Moreover, the experimental evaluation indicates that the use of the proposed representations provides stable classifiers with strong quantitative results, especially in semantically-complex classification scenarios.

Over the years, quite a lot of research has been done on using different machine learning techniques for Face Recognition (FR) for identifying the faces of different people. Current FR techniques are still not accurate enough for real world scenarios and pose a lot of problems against varying illumination levels, pose variations, noise and occlusion in the image. Thus, a single keypoint extraction technique may not be suitable for all cases. Hence, in this paper a novel technique is proposed for Keypoint Fusion (KF) obtained by fusing SIFT, SURF and ORB keypoints which is more accurate and suitable for real time application. The paper is also focused on proposing a novel technique of using a Self-Organizing Map (SOM) and Vector of Locally Aggregated Descriptors (VLAD) for image clustering. VLAD is used to extend the SOM’s ability to cluster keypoint descriptors. Image classification is carried out using a SGD (Stochastic Gradient Descent) based SVM (Support Vector Machine) classifier. The performance of classification of the proposed framework on benchmark datasets (Grimace, Faces95 and Faces96) has been tabulated and compared with other standard techniques. It is seen that the proposed framework performs better than the BOW (Bag Of Words) model and the KF technique was accurate and quick enough to beat the traditional keypoint extraction techniques.

We analyse the environment scene classification methods based on the Bag of Words (BoW) model. The BoW model encodes images by a bag of visual features, which is a sparse histogram over a dictionary of visual features extracted from an image. We analyse five feature detectors (Scale Invasive Feature Transform (SIFT), Speed-Up Robust Features (SURF), Features from Accelerated Segment Test (FAST), Maximally Stable Extremal Regions (MSER), and grid-based) and three feature descriptors (SIFT, SURF and U-SURF). Our experiments show that feature detection with a grid and feature description using SIFT descriptor, and feature detection with SURF and feature description with U-SURF are most effective when classifying (using Support Vector Machine (SVM)) images into eight outdoor scene categories (coast, forest, highway, inside city, mountain, open country, street, and high buildings). Indoor scene classification into five categories (bedroom, industrial, kitchen, living room, and store) achieved worse results, while the most confused categories were industrial/store images. The classification of full image dataset (15 outdoor and indoor categories) achieved the overall accuracy of 67.49 ± 1.50%, while most errors came from misclassifications of indoor images. The results of the study can be applicable for assisting living applications and security systems.

A specific word can have different meanings depending on the context in which it appears. Identifying the proper sense of the word is crucial in many tasks such as Machine Translation, Anaphora Resolution, Search Engine Recommendation. A word with wrong sense can make the whole sentence meaningless. The task of Word Sense Disambiguation (WSD), is to assign a sense to the ambiguous word based on the context. This is a basic problem which needs to be resolved in the field of NLP and has a variety of solutions. Usual approaches involve supervised machine learning techniques which uses bag-of-words approach. But to achieve better results, WSD should move to sequence modeling rather than the traditional bag of words approach for better performance. Deep neural networks have been used for variety of NLP tasks and a Bi-Directional Long Short-Term Memory (BiLSTM) or a Bi-Directional Gated Recurrent Unit (BiGRU) are possible candidate solutions. The proposed model uses one word per BiGRU or BiLSTM approach, where each word has a model trained for disambiguation. This helps in easy modification of existing model whenever needed. A comparative study on various deep learning models is also performed. The evaluation study of the models shows Bi-Directional models outperforming the other deep learning models.