Enhancing enterprise knowledge processes via cross-media extraction

Topic modeling continues to grow as a popular technique for finding hidden patterns, as well as grouping collections of new types of text and non-text data. Recent years have witnessed a growing body of work in developing metrics and techniques for evaluating the quality of topic models and the topics they generate. This is particularly true for text data where significant attention has been given to the semantic interpretability of topics using measures such as coherence. It has been shown however that topic assessments based on coherence metrics do not always align well with human judgment. Other efforts have examined the utility of information-theoretic distance metrics for evaluating topic quality in connection with semantic interpretability. Although there has been progress in evaluating interpretability of topics, the existing intrinsic evaluation metrics do not address some of the other aspects of concern in topic modeling such as: the number of topics to select, the ability to align topics from different models, and assessing the quality of training data. Here we propose an alternative metric for characterizing topic quality that addresses all three aforementioned issues. Our approach is based on clustering topics, and using the silhouette measure, a popular clustering index, for characterizing the quality of topics. We illustrate the utility of this approach in addressing the other topic modeling concerns noted above. Since this metric is not focused on interpretability, we believe it can be applied more broadly to text as well as non-text data. In this paper however we focus on the application of this metric to archival and non-archival text data.

In today’s technology world, data security has a great importance. Because each data type has its own characteristics, there are various methods of providing this security. The main subject of this study is the security of image data which are more complex structures than text data. Using traditional encryption methods alone to ensure security in image data types can create security weaknesses. For this reason, nowadays, some traditional methods are combined with each other or different methods to encrypt image data. In this study, 131 articles were examined and image encryption methods were classified according to the traditional methods they contain. Studies on both colored and gray level images have been handled together. Finally, the images used in the articles were compared with each other in many ways and the results were presented graphically.

Using accurate and high quality data is critical for any application relying heavily on the data, be it machine learning, artificial intelligence, or digital twins. Bad quality and erroneous data can result in inaccurate predictions even if the model is otherwise robust. Ensuring data quality is more critical in real-time applications where there is no human in the loop to perform sense checks on data or results. A real-time digital twin implementation for a floating system uses time-series data from numerous measurements such as wind, waves, GPS, vessel motions, mooring tensions, draft, etc. Statistics computed from the data are used in the digital twin. An extensive data checking and cleaning routine was written that performs data quality checks and corrections on the time series data before statistics are computed. Various types of errors that typically occur in a time series include noise, flat-lined data, clipped data, outliers, and discontinuities. Statistical procedures were developed to check the raw time-series for all these errors. The procedures are generic and robust so they can be used for different types of data. Some data types are slow varying (e.g., GPS) while the others are fast varying random processes. A measurement classified as an error in one type of data is not necessarily an error in the other data type. For example, GPS data can be discontinuous by nature but a discontinuity in the wave data indicates an error. Likewise, checking for white noise in mooring tension data is not that meaningful. We developed parametric data procedures so that the same routine can handle different types of data and their errors. Outlier removal routines use the standard deviation of the time-series which itself could be biased from errors. Therefore, a method to compute unbiased statistics from the raw data is developed and implemented for robust outlier removal. Extensive testing on years of measured data and on hundreds of data channels was performed to ensure that data cleaning procedures function as intended. Statistics (mean, standard deviations, maximum, and minimum) were computed from both the raw and cleaned data. Comparison showed significant differences in raw and cleaned statistics, with the latter obviously being more accurate. Data cleaning, while not sounding as high tech as other analytics algorithms, is a critical foundation of any data science application. Using cleaned time-series data and corresponding statistics ensure that a data analytics model provides actionable results. Clean data and statistics help achieve the intended purpose of the digital twin, which is to inform operators of the health/condition of the asset and flag any anomalous events.

Data is said to be the new oil. This applies to textual data as well. The amount of plain textual data on the Internet is abundant, and it is still growing. However, similar to oil, raw data is not valuable by itself. Its value and potential are availed through further accurate processing and analysis. And this is where Natural Language Processing (NLP) and its procedures come into the picture. Since NLP tasks require standardized and high-quality inputs for high efficiency and performance, raw textual data requires at least a basic level of cleaning and standardization. Therefore, text normalization - a process of transforming noisy data into improved, i. e., a standard representation - is often a prerequisite for a variety of NLP tasks, including information extraction, machine translation, or sentiment analysis. The Persian language, being the 5th content language of the Web, can be a great source of diverse information. However, several language-specific characteristics hinder the potential application of Persian data. The focus of this chapter lies in describing challenges that the Persian language poses for NLP and evaluating the impact of normalizing a large Persian corpus for one of the downstream NLP tasks - the discovery of multiword expressions (MWEs).

Stacked isomorphic autoencoder based soft analyzer and its application to sulfur recovery unit

We address the problem, fundamental to linguistics, bioinformatics, and certain other disciplines, of using corpora of raw symbolic sequential data to infer underlying rules that govern their production. Given a corpus of strings (such as text, transcribed speech, chromosome or protein sequence data, sheet music, etc.), our unsupervised algorithm recursively distills from it hierarchically structured patterns. The adios (automatic distillation of structure) algorithm relies on a statistical method for pattern extraction and on structured generalization, two processes that have been implicated in language acquisition. It has been evaluated on artificial context-free grammars with thousands of rules, on natural languages as diverse as English and Chinese, and on protein data correlating sequence with function. This unsupervised algorithm is capable of learning complex syntax, generating grammatical novel sentences, and proving useful in other fields that call for structure discovery from raw data, such as bioinformatics.

There are some problems in the active participation of college students in ideological and mental health support services in China, such as low attention, low participation, and high data redundancy. Based on this, this paper studies the active participation of college students' ideological and mental health support service based on fuzzy cluster analysis algorithm. Compared with the disadvantages of the current mainstream discrete optimization analysis models on mental health (such as high-dimensional enterprise model, Dajiaweikang model, and short-range group control model), which need to set the known data gradient interval, this paper creatively adopts the fuzzy cluster analysis algorithm, based on the characteristics of different types of college students' ideological and mental health problems. Combined with the improved star discrete analysis model, this paper constructs the active participatory evaluation strategy of college students' ideological and mental health support services. On this basis, the model can not only record and store the participatory data of ideological and mental health support for students of different grades but also match and track different types of data based on special framework conditions, so as to achieve numerical normal analysis and directional matching for the data coupling mode of college students' ideological and mental health support services. On the other hand, the Planck constant factor is used to classify different types of ideological and psychological factor data, and combined with the idea of fuzzy clustering, the hierarchical analysis and quantitative calibration of different types of data groups are realized, so as to improve the reliability and authenticity of the active participation in college students' mental health support services. The results show that this star discrete analysis model can analyze the active participation of college students' ideological and mental health support services according to the data matching degree of different levels and can effectively improve the analysis efficiency of data vectors. Compared with the traditional research methods on the active participation of college students' ideological and mental health support services, this method can realize the matching and tracking of different types of data, so as to make a numerical and normal analysis on the data coupling mode of college students' ideological and mental health support services.

Focusing on emotion recognition, this paper addresses the task of emotion classification and its performance with respect to accuracy, by investigating the capabilities of a distributed ensemble model using precision-based weighted blending. Research on emotion recognition and classification refers to the detection of an individual’s emotional state by considering various types of data as input features, such as textual data, facial expressions, vocal, gesture and physiological signal recognition, electrocardiogram (ECG) and electrodermography (EDG)/galvanic skin response (GSR). The extraction of effective emotional features from different types of input data, as well as the analysis of large volume of real-time data, have become increasingly important tasks in order to perform accurate classification. Taking into consideration the volume and variety of the examined problem, a machine learning model that works in a distributed manner is essential. In this direction, we propose a precision-based weighted blending distributed ensemble model for emotion classification. The suggested ensemble model can work well in a distributed manner using the concepts of Spark’s resilient distributed datasets, which provide quick in-memory processing capabilities and also perform iterative computations effectively. Regarding model validation set, weights are assigned to different classifiers in the ensemble model, based on their precision value. Each weight determines the importance of the respective classifier in terms of its performing prediction, while a new model is built upon the derived weights. The produced model performs the task of final prediction on the test dataset. The results disclose that the proposed ensemble model is sufficiently accurate in differentiating between primary emotions (such as sadness, fear, and anger) and secondary emotions. The suggested ensemble model achieved accuracy of 76.2%, 99.4%, and 99.6% on the FER-2013, CK+, and FERG-DB datasets, respectively.

Stock markets are at the heart of market economies, and stock market prediction is a hot topic in research. There are many methods of prediction, and they often use different types of data. The two main types are technical and fundamental data. This paper explores company-specific prediction through predicting the direction of the Apple stock. Three neural networks that use different input data are compared. The Tech model uses technical data, the Comp model uses numerical fundamental data with an emphasis on company data, and the Text model uses textual news data. Several metrics, including ones useful for preventing bias when dealing with imbalanced data, were used to compare the models. The Comp and Text models showed prediction bias, and two more models, W-Comp and W-Text, were trained to attempt to mitigate that bias. The Comp model performed the best out of the original models, and W-Comp also showed good performance. W-Text performed the best out of all the models. This suggests that numerical fundamental data is useful in predicting the market, and textual data also has potential. Future research could be used to improve the performance of all the models and more thoroughly compare the types of data used.

A Comparative Study of Outfit Recommendation Methods with a Focus on Attention-based Fusion

Computed tomography (CT) is of great significance in the field of medical diagnosis. However, metal artifacts in the reconstruction images are disadvantageous for doctors to make a fast and accurate diagnosis when high-density metals present in the scanned location. The spectral CT has excellent performance in metal artifact reduction (MAR) method, which can combinate the prior information can to realize the information complementarity. In this paper, a MAR method based on dual-energy raw data constrained learning is proposed in this paper. The raw projection data of high/low energy and the results of normalized metal artifact reduction (NMAR) are input to the dual-stream U-Net (DSU-Net) for getting the virtual monoenergetic image (VMI) to reduce the secondary artifacts. The experimental results show that the peak signal-to-noise ratio (PSNR) of the output image is up to 49.60, SSIM to 0.997. It is proved that the raw data constrained learning method can suppress the residual artifacts from the traditional information pretreatment method.

Fast generation of weather-radar echo maps in vector format and accurate localization of weighted centroid of dBZ (dBZ stands for decibels of reflectivity of a radar signal reflected off a remote object) are the basis of studying the characteristic tracking algorithms which are based on the vector echoes. The authors principally studied the approach to generating the vector echo map, and discussed the localization methods of weighted centroid of dBZ on vector echo maps. First, based on the traditional calculation method on raster echo data, some new localization methods of weighted centroid of dBZ on vector echo data were proposed by considering the weights of features’ area and distance from their location to radar center. Second, taking the base reflectivity products of CINRAD/SA weather radar in Meizhou city of China as data sources, they illustrated the storage structure of this type of echo data and studied the drawing mode of changing this type of data into vector format files under the polar coordinate system in detail. Third, using the same vector echo maps created by the above method, the weighted centroid of the same area was calculated by the above localization methods. In the end, Compared with the calculated value of the same area by traditional method which is based on raster echo maps, the three new calculated results and the sources of error were analyzed in detail and two conclusions were drawn: the echo’s precision in vector format is much higher than that in raster format and it is more accurate to take the features’ area and distance to radar center as weights during the calculation of weighted centroid of dBZ on echo maps in vector format. DOI: http://dx.doi.org/10.11591/telkomnika.v11i2.2071

Deep learning algorithms can be used to design systems that report data on patients and deliver warnings to medical applications or electronic health information if there are changes in the patient's health. These systems could be created using deep Learning. This helps verify that patients get the proper effective care at the proper time for each specific patient. A healthcare decision support system was presented using the Internet of Things and deep learning methods. In the proposed system, we examined the capability of integrating deep learning technology into automatic diagnosis and IoT capabilities for faster message exchange over the Internet. We have selected the suitable Neural Network structure (number of best-hidden layers and activation function classes) to construct the e-health system. In addition, the e-health system relied on data from doctors to understand the Neural Network. In the validation method, the total evaluation of the proposed healthcare system for diagnostics provides dependability under various patient conditions. Based on evaluation and simulation findings, a dual hidden layer of feed-forward NN and its neurons store the tanh function more effectively than other NN. To overcome challenges, this study will integrate artificial intelligence with IoT. This study aims to determine the NN's optimal layer counts and activation function variations.

The real-time system of the JT-60 has a cascade structure with the major and the minor control loops. The multi-variable feedback control is adopted in the major control loop in order to adjust non-linear effects which will appear in the high-/beta/region. By applying the concept of 'phase shift control' to the control program of the major control loop, a flexible tokamak discharge control is expected to be performed with taking account of machine protection.

The data integration time of traditional enterprise employee information system data integration method is long, and the data coverage rate is not high in the integration process. This paper designs a data integration method of enterprise employee information system based on big data technology. First of all, the unified enterprise information system data integration standards and data classification principles are set. According to the relationship between data, data types are defined, employee information system data are classified, and inconsistent and duplicate data are eliminated after classification, so as to reduce the interference of invalid data and improve the efficiency of data integration. Then, a unified coding standard is formulated to check and code the classified data to determine whether the classified data set matches the actual data. Based on the offline data, deep learning algorithm is used to integrate the data of employee information system, which can reduce the occupancy rate of resources in the system and improve the coverage of effective data. The experimental results show that the data integration method takes less time than the traditional method, and has high data coverage, so it has practical application significance.