Representation Techniques Research Articles

Financial sentiment analysis: Classic methods vs. deep learning models

Sentiment Analysis, also known as Opinion Mining, gained prominence in the early 2000s alongside the emergence of internet forums, blogs, and social media platforms. Researchers and businesses recognized the imperative to automate the extraction of valuable insights from the vast pool of textual data generated online. Its utility in the business domain is undeniable, offering actionable insights into customer opinions and attitudes, empowering data-driven decisions that enhance products, services, and customer satisfaction. The expansion of Sentiment Analysis into the financial sector came as a direct consequence, prompting the adaptation of powerful Natural Language Processing models to these contexts. In this study, we rigorously test numerous classical Machine Learning classification algorithms and ensembles against five contemporary Deep Learning Pre-Trained models, like BERT, RoBERTa, and three variants of FinBERT. However, its aim extends beyond evaluating the performance of modern methods, especially those designed for financial tasks, to a comparison of them with classical ones. We also explore how different text representation and data augmentation techniques impact classification outcomes when classical methods are employed. The study yields a wealth of intriguing results, which are thoroughly discussed.

Fake news detection: Taxonomy and comparative study

The proliferation of social networks has presented a significant challenge in combating the pervasive issue of fake news within modern societies. Due to the large amount of information and news produced daily in text, audio, and video, the validation and verification of this information have become crucial tasks. Leveraging advancements in artificial intelligence, distinguishing between fake news and factual information through automatic fake news detection systems has become more feasible. Automatic fake news detection has been explored from diverse perspectives, employing various feature extraction and classification models. Nonetheless, empirical evaluations, categorization, and comparisons of existing techniques for handling this problem remain limited. In this paper, we revisit the definitions and perspectives of fake news and propose an updated taxonomy for the field based on multiple criteria: (1) Type of features used in fake news detection; (2) Fake news detection perspectives; (3) Feature representation methods; and (4) Classification approaches. Moreover, we conduct an extensive empirical study to evaluate several feature representation techniques and classification approaches based on accuracy and computational cost. Our experimental results demonstrate that the optimal feature extraction techniques vary depending on the characteristics of the dataset. Notably, context-dependent models based on transformer models consistently exhibit superior performance. Additionally, employing transformer models as feature extraction methods, rather than solely fine-tuning the network for the downstream task, improves overall performance. Through extensive error analysis, we identify that a combination of feature representation methods and classification algorithms, including classical ones, offer complementary aspects and should be considered for achieving better generalization performance while maintaining a relatively low computational cost. For further details, including source codes, figures, and datasets, please refer to our project’s GitHub repository: [https://github.com/FFarhangian/Fake-news-detection-Comparative-Study].

The Impact of Data on Structure-Based Binding Affinity Predictions Using Deep Neural Networks.

Artificial intelligence (AI) has gained significant traction in the field of drug discovery, with deep learning (DL) algorithms playing a crucial role in predicting protein-ligand binding affinities. Despite advancements in neural network architectures, system representation, and training techniques, the performance of DL affinity prediction has reached a plateau, prompting the question of whether it is truly solved or if the current performance is overly optimistic and reliant on biased, easily predictable data. Like other DL-related problems, this issue seems to stem from the training and test sets used when building the models. In this work, we investigate the impact of several parameters related to the input data on the performance of neural network affinity prediction models. Notably, we identify the size of the binding pocket as a critical factor influencing the performance of our statistical models; furthermore, it is more important to train a model with as much data as possible than to restrict the training to only high-quality datasets. Finally, we also confirm the bias in the typically used current test sets. Therefore, several types of evaluation and benchmarking are required to understand models' decision-making processes and accurately compare the performance of models.

Interval Type-2 Fuzzy Relation Matrix Model via Semitensor Product

A new modeling idea of interval type-2 fuzzy logic system (IT2 FLS) is proposed on the basis of matrix semi-tensor product (STP) fuzzy representation technique, to construct an interval type-2 fuzzy relationship matrix (IT2 FRM) model. The model naturally inherits the advantages of STP theory, which enables the fuzzy logic reasoning process to be translated into algebraic form for computation. In most cases, the fuzzy rules in the IT2 FRM are generally unknown. To address the above situation, the parameters of this model can be trained with observed data pairs. Firstly, the fuzzy rules are generated by the space division method, and the initial parameters of the model are determined. Secondly, the parameters in the model are trained using the steepest descent algorithm. Finally, the IT2 FRM model is validated by simulation, and the results demonstrate that the IT2 FLS can be designed using the STP method.

CARMEL: Capturing spatio-temporal correlations via time-series sub-window imaging for home appliance classification

Energy management systems (EMS), as enablers of more efficient energy consumption, monitor and manage appliances to help residents be more energy efficient and thus more frugal. Recent appliance detection and identification techniques for such systems rely on machine learning. However, machine learning solutions for appliance classification on existing low-frequency household metering have not yet been thoroughly investigated. In this paper, we propose CARMEL, a new approach for identifying home appliances from load monitoring in building EMS based on a new data representation technique and a new model that leverages spatio-temporal correlations in the new representation. The proposed data representation technique performs dimensionality expansion of time series that scales linearly, rather than quadratically and, together with the proposed model, outperform the state of the art image transformation models by 5 percentage points. Evaluation on 5 different low-frequency household metering datasets, considering 29 appliances in total, shows that the proposed representation and the corresponding resource-aware deep learning architecture (1) achieve an average weighted F1 score of 0.92 and (2) require only 230 labeled samples and 3x fewer epochs to transfer to new households.

Note on the Classification of the Unitary $\rm{SL}_2(\mathbb{R})$ Representations

The aim of this paper is to explain the classification of the unitary SL_2(R) representations done by Gelfand by using the induced representation technique. We induce the SL_2(R) representation from the subgroup N. We get a representation constructed on a space of homogeneous functions in two variables. Then, we move to induce the SL_2(R) representation in stages. Consequently, the representation of SL_2(R) acts on a space of functions of one variable.

Generalized decision directed acyclic graphs for classification tasks

The recent studies present the various perspectives of Formal Concept Analysis from the Artificial Intelligence point of view since it can provide powerful techniques for knowledge representation, classification tasks, recommendation techniques, or data analysis. Formal Concept Analysis can be generally seen as an unsupervised Machine Learning technique that applies mathematical lattice theory to organize data based on objects and their shared attributes. However, the connections of Formal Concept Analysis with supervised learning are also thoroughly investigated. Regarding supervised learning, the learning algorithms of random forests or decision jungles have become popular and powerful Machine Learning ensemble methods. In this paper, we propose a generalization of a classification method based on the generalized decision directed acyclic graphs. We outline and review the recent studies on connections between Formal Concept Analysis and supervised learning. Finally, we present the results of our experiments regarding several datasets from the UCI Machine Learning repository.

Beyond the Meta: Leveraging Game Design Parameters for Patch-Agnostic Esport Analitics

Esport games comprise a sizeable fraction of the global games market, and is the fastest growing segment in games. This has given rise to the domain of esports analytics, which uses telemetry data from games to inform players, coaches, broadcasters and other stakeholders. Compared to traditional sports, esport titles change rapidly, in terms of mechanics as well as rules. Due to these frequent changes to the parameters of the game, esport analytics models can have a short life-spam, a problem which is largely ignored within the literature. This paper extracts information from game design (i.e. patch notes) and utilises clustering techniques to propose a new form of character representation. As a case study, a neural network model is trained to predict the number of kills in a Dota 2 match utilising this novel character representation technique. The performance of this model is then evaluated against two distinct baselines, including conventional techniques. Not only did the model significantly outperform the baselines in terms of accuracy (85% AUC), but the model also maintains the accuracy in two newer iterations of the game that introduced one new character and a brand new character type. These changes introduced to the design of the game would typically break conventional techniques that are commonly used within the literature. Therefore, the proposed methodology for representing characters can increase the life-spam of machine learning models as well as contribute to a higher performance when compared to traditional techniques typically employed within the literature.

John Fowles’s diaries as a creative biography of the writer

J. Fowles’s diaries illustrate the emergence of the creative career of the prominent 20th-century English writer. In Russian science, there is an opinion that Fowles’s oeuvre belongs to postmodernism. However, Fowles persistently separated himself from postmodernism. A thorough study of his diaries testifies that with all the variety of themes and images and the abundance of the techniques of artistic representation of reality, the key issue of his works was the theme of the individual gaining self-awareness as a necessary condition for achieving freedom. His diaries can be considered as a compact documentary and fictional work, as a presentation of the principles of literary development and its continuity. Fowles’s diary is a peculiar confession of a young man, his creative pursuits and doubts of self–determination in society and life. In his aesthetic principles Fowles focuses on the classical tradition of literature in the 18th–19th centuries. Continuing the tradition of Rousseau, Fowles creates a model of a man who is characterized by the cult of emotions and nature. For Fowles, the world of art is the world of spirituality. Fowles’s diaries allow us to consider the attitude of the autobiographical protagonist to important phenomena of public life. In his diaries, special attention is paid to the topics of time, memory and nature. The diary of John Fowles describes the process of the person’s psychological formation with a real chronotope, with an analytical, informative and aesthetically rich word.

The effect of movement representation techniques on ankle function and performance in persons with or without a lateral ankle sprain: a systematic review and meta-analysis

BackgroundLateral ankle sprains are highly prevalent and result in tissue damage, impairments of muscle strength, instability, and muscle activation. Up to 74% will experience ongoing symptoms after a lateral ankle sprain. In healthy subjects, motor imagery might induce neural changes in the somatosensory and motor areas of the brain, yielding favourable enhancements in muscular force. However, during motor imagery, difficulties in building a motor image, no somatosensory feedback, and the absence of structural changes at the level of the muscle might explain the differences found between motor imagery and physical practice. In rehabilitation, motor imagery might be supportive in rebuilding motor networks or creating new networks to restore impairments in muscle activation and movement patterns. This systematic review was undertaken to summarize the current body of evidence about the effect on motor imagery, or action observation, on lower leg strength, muscle performance, ankle range of motion, balance, and edema in persons with, and without, a lateral ankle sprain compared to usual care, a placebo intervention, or no intervention.MethodsA systematic review with meta-analysis of randomized controlled trials was conducted in healthy participants and participants with a lateral ankle sprain. Motor imagery or action observation in isolation, or in combination with usual care were compared to a placebo intervention, or no intervention. An electronic search of MEDLINE, EMBASE, Cinahl, Psychinfo, Sportdiscus, Web of Science, Cochrane and Google Scholar was conducted, and articles published up to 7th June 2023 were included. Two reviewers individually screened titles and abstracts for relevancy using the inclusion criteria. Variables related to muscle strength, muscle function, range of motion, balance, return to sports tests, or questionnaires on self-reported function or activities were extracted. A risk of bias assessment was done using the Cochrane Risk-of-Bias tool II by two reviewers. Meta-analysis using a random effects model was performed when two or more studies reported the same outcome measures. The Standardized Mean Difference (SMD) was calculated over the change from baseline scores. Review manager 5.4 was used to perform analysis of subgroup differences and test for statistically significant differences. Confidence intervals were visually checked for overlap between subgroups.ResultsNine studies, six examining healthy participants and three examining participants with an acute lateral ankle sprain, were included. All studies were rated with moderate to high risk of bias overall. Quality of the motor imagery interventions differed largely between studies. Meta-analysis showed a large and significant effect of motor imagery on lower leg strength (SMD 1.47, 95% CI 0.44 to 2.50); however, the evidence was downgraded to very low certainty due to substantial heterogeneity (I2 = 73%), limitations in the studies (some concerns in risk of bias in all studies), and imprecision (n = < 300). Evidence showed no association with ankle range of motion (SMD 0.25, 95% CI -0.43 to 0.93), edema (SMD -1.11, 95% CI -1.60 to 3.81), the anterior reach direction of the Star Excursion Balance Test (SEBT) (SMD 0.73, 95% CI -0.62 to 2.08), the posterolateral direction (SMD 0.32, 95% CI -0.94 to 1.57), and the posteromedial direction (SMD 0.52, 95% CI -0.07 to 1.10). The certainty of evidence for the different comparisons was very low.ConclusionsThere is a low certainty, significant, positive effect for motor imagery being able to improve lower leg muscle strength in healthy participants. The effect on balance, range of motion and edema was uncertain and of very low certainty.Systematic review registrationPROSPERO CRD42021243258.

Synergizing Low Rank Representation and Deep Learning for Automatic Pavement Crack Detection

Due to the critical role of pavement crack detection for road maintenance and eventually ensuring safety, remarkable efforts have been devoted to this research area, and such a trend is further intensified for the coming unmanned vehicle era. However, such crack detection task still remains unexpectedly challenging in practice since the appearance of both cracks and the background are diverse and complex in real scenarios. In this work, we propose an automatic pavement crack detection method via synergizing low rank representation (LRR) and deep learning techniques. First, leveraging LRR which facilitates anomaly detection without making any specific assumption, we can easily discriminate most of the frames with cracks from the long sequence with a consistent pavement base, followed by a straightforward algorithm to localize the cracks. In order to achieve the intelligence of detecting cracks with different pavement basis under unconstrained imaging conditions, we resort to deep learning techniques and propose a deep convolutional neural network for crack detection leveraging on multi-level features and atrous spatial pyramid pooling (ASPP). We train this network based on the training data obtained in the previous stage in an end-to-end manner. Extensive experiments on a wide range of pavements demonstrate the high performance in terms of both accuracy and automaticity. Moreover, the dataset generated by us is much more extensive and challenging than public ones. We put it online at https://gaozhinuswhu.com to benefit the community.

A Secure and Efficient Biometric Template Protection Scheme for Palmprint Recognition System

The accuracy and security of a biometric system are the two sides of a coin. A biometric system must be simple, flexible, efficient, and secure enough from unauthorized access. Concerning these requirements, this paper presents an enhanced palmprint recognition system with superior performance and improved biometric template protection schemes. This method comprises five components: image preprocessing, feature extraction, cancelable biometrics, classification, and Bio-cryptosystem. The palm texture region is extracted from the input hand image during preprocessing. A feature representation technique is then used to extract discriminant features from the palmprint texture patterns. The extracted features are then analyzed using an information coding scheme to compute a user-specific token. Here, the user-specific tokens along with the original palmprint features are used to calculate the feature vectors for the cancelable biometric system. A novel Bio-Cryptosystem has been proposed to protect templates with high security where there is no need to remember the users&#x2019; tokens. Cancelable feature vectors are kept online in encrypted form using the proposed Bio-Cryptosystem, which is used to identify or verify the subjects after decryption. The original feature vectors are kept offline to prevent these from external attacks and misuse. Finally, the multi-class linear support vector machine has been used for user classification. For experimental purposes, two benchmark hand image databases: BOSPHORUS and CASIA-Palmprint, are employed, and the proposed system has achieved nearly 100&#x0025; correct recognition rate in cancelable domain for both the employed databases. Finally, the performance comparison in terms of both verification and identification for the palmprint and the cancelable palmprint recognition systems with the existing state-of-the-art methods shows the superiority of the proposed system. <i>Impact Statement</i> &#x2014;We have improved the biometric template protection scheme by modifying the cancelable biometric technique with two security levels capable of protecting the biometric information from the dictionary, brute-force, database, channel, replay, and spoofing attacks. To avoid the difficulty of remembering long user passwords/tokens, an automatic key generation technique of user passwords has been proposed. Two different keys are generated: the first one used for cancelable biometrics and the second for encryption-decryption processes. Both these keys are not needed to store or remember and are helpful for a multi-factor authentication system. A novel Bio-cryptographic algorithm instead of traditional Bio-Cryptographic algorithms have been introduced within the proposed system. Several attacks and challenging issues of biometric recognition systems have been considered with maintaining outstanding performance in terms of accuracy and security.

Resource-Aware Time Series Imaging Classification for Wireless Link Layer Anomalies.

The number of end devices that use the last-mile wireless connectivity is dramatically increasing with the rise of smart infrastructures and requires reliable functioning to support smooth and efficient business processes. To efficiently manage such massive wireless networks, more advanced and accurate network monitoring and malfunction detection solutions are required. In this article, we perform a first-time analysis of image-based representation techniques for wireless anomaly detection using recurrence plots (RPs) and Gramian angular fields and propose a new deep learning architecture enabling accurate anomaly detection. We elaborate on the design considerations for developing a resource-aware architecture and propose a new model using time series to image transformation using RPs. We show that the proposed model: 1) outperforms the one based on Gramian angular fields by up to 14% points; 2) outperforms classical ML models using dynamic time warping by up to 24% points; 3) outperforms or performs on par with mainstream architectures, such as AlexNet and VGG11 while having <10x their weights and up to ≈ 8% of their computational complexity; and d) outperforms the state of the art in the respective application area by up to 55% points. Finally, we also explain on randomly chosen examples how the classifier takes decisions.

Two-Stage Sparse Representation Clustering for Dynamic Data Streams.

Data streams are a potentially unbounded sequence of data objects, and the clustering of such data is an effective way of identifying their underlying patterns. Existing data stream clustering algorithms face two critical issues: 1) evaluating the relationship among data objects with individual landmark windows of fixed size and 2) passing useful knowledge from previous landmark windows to the current landmark window. Based on sparse representation techniques, this article proposes a two-stage sparse representation clustering (TSSRC) method. The novelty of the proposed TSSRC algorithm comes from evaluating the effective relationship among data objects in the landmark windows with an accurate number of clusters. First, the proposed algorithm evaluates the relationship among data objects using sparse representation techniques. The dictionary and sparse representations are iteratively updated by solving a convex optimization problem. Second, the proposed TSSRC algorithm presents a dictionary initialization strategy that seeks representative data objects by making full use of the sparse representation results. This efficiently passes previously learned knowledge to the current landmark window over time. Moreover, the convergence and sparse stability of TSSRC can be theoretically guaranteed in continuous landmark windows under certain conditions. Experimental results on benchmark datasets demonstrate the effectiveness and robustness of TSSRC.

How should we treat painful sensitivity in the hand? An international e-Delphi study

BackgroundEvidence synthesis suggests allodynia resulting from neuropathic pain has few interventions with clear effectiveness. As research continues to build this needed evidence base, expert consensus recommendations can address the conflicting approaches within current hand therapy practice. PurposeThis study aimed to develop consensus recommendations for the clinical management of allodynia from an international panel of hand therapists. Study DesignThis was an international e-Delphi survey study. MethodsWe recruited international hand rehabilitation experts to participate in an e-Delphi survey. Consensus was defined as 75% or more of participants agreeing with a recommendation, and at least 3 rounds of consensus building were anticipated. Experts were identified from 21 countries, and clinical vignettes describing a spectrum of patients with painful sensitivity in the hand were provided to elicit treatment recommendations. Initial recommendations were summarized, and consensus sought for clinical practice recommendations. ResultsSixty-eight participants were invited, with 44 more added through peer nominations. Fifty-four participants from 19 countries completed the initial survey and were invited to participate in all subsequent rounds. Over 900 treatment suggestions were provided from the initial vignettes across domains, including sensory, physical, and functional interventions, education, and cortical representation techniques: 46 ultimately reached consensus. However, important discrepancies in justification (eg, why allodynia should be covered) and implementation of techniques (eg, desensitization, sensory reeducation) were identified as the consensus exercise progressed. ConclusionsExperts recommend individually tailored programs to treat allodynia using a variety of physical/movement, sensory-based, and “top-down” approaches; this is highly aligned with contemporary theories, such as the Neuromatrix Model of Pain. However, consensus was not reached on the justification and implementation of some of these approaches, reflecting the lack of a taxonomy and supporting evidence for tactile stimulation approaches in the current literature. Trials directly comparing the effectiveness of these approaches are needed.

SDbQfSum: Query‐focused summarization framework based on diversity and text semantic analysis

AbstractQuery‐focused multi‐document summarization (Qf‐MDS) is a sub‐task of automatic text summarization that aims to extract a substitute summary from a document cluster of the same topic and based on a user query. Unlike other summarization tasks, Qf‐MDS has specific research challenges including the differences and similarities across related document sets, the high degree of redundancy inherent in the summaries created from multiple related sources, relevance to the given query, topic diversity in the produced summary and the small source‐to‐summary compression ratio. In this work, we propose a semantic diversity feature based query‐focused extractive summarizer (SDbQfSum) built on powerful text semantic representation techniques underpinned with Wikipedia commonsense knowledge in order to address the query‐relevance, centrality, redundancy and diversity challenges. Specifically, a semantically parsed document text is combined with knowledge‐based vectorial representation to extract effective sentence importance and query‐relevance features. The proposed monolingual summarizer is evaluated on a standard English dataset for automatic query‐focused summarization tasks, that is, the DUC2006 dataset. The obtained results show that our summarizer outperforms most state‐of‐the‐art related approaches on one or more ROUGE measures achieving 0.418, 0.092 and 0.152 in ROUGE‐1, ROUGE‐2, and ROUGE‐SU4 respectively. It also attains competitive performance with the slightly outperforming system(s), for example, the difference between our system's result and best system in ROUGE‐1 is just 0.006. We also found through the conducted experiments that our proposed custom cluster merging algorithm significantly reduces information redundancy while maintaining topic diversity across documents.

A Novel Fuzzy Unsupervised Feature Learning Approach

The effectiveness of machine learning approaches depends on the quality of the data representation. In fact, some representations may mislead such learning approaches upon concealing relevant explanatory variables. Although feature engineering, that utilizes domain knowledge and/or expert supervision, yields typical data representation techniques, generic unsupervised feature learning represents an even more objective alternative to determine relevant attributes and generate optimal feature spaces. In this paper, we propose a new fuzzy unsupervised feature learning approach (FUL) that automatically derives new features by revealing the intrinsic structure of the data. In fact, FUL exploits the clusters and the associated fuzzy memberships generated by a fuzzy C-means algorithm, and devises new basis functions and their corresponding representation. The experiments results showed that FUL overtakes relevant state of the art approaches. It yielded the highest F1-measure with an improvement of 8%, 11%, 3%, and 4% on Parkinson, Epilepsy, Gait, and breast cancer datasets, respectively.

Research on Emotional Analysis of Online Book Reviews Based on Word2Vec Method

At present, emotional analysis in the field of natural language processing is gradually becoming a popular research direction on social media. With the rise of artificial intelligence technology and deep learning algorithms, word vector representation methods based on neural networks have emerged and emerged in the field of sentiment analysis. According to the current research results of experts, the Word2Vec model has become the mainstream word vector representation method. The algorithm model is based on continuous optimization and iteration of random initialization word vectors, ultimately obtaining a stable performance word vector representation. If the research goal is to obtain sentence vectors, weighting processing is also required. Therefore, after systematically summarizing and analyzing the representation techniques of initialized word vectors and solving sentence vectors through word vector weighting, this article proposes corresponding improvement methods. This article improves the initialization algorithm of its word vector based on the Word2Vec model. On the basis of the original randomly generated word vector, the correlation information of words is added to change the distribution characteristics of the initial word vector, thereby enhancing the performance of the word vector. In the comparative experiment of the user comment dataset, the improved model showed a 1% to 3% improvement in sentiment polarity classification accuracy compared to the other model.

Harnessing Deep Learning Techniques for Text Clustering and Document Categorization

This research paper delves into the realm of deep text clustering algorithms with the aim of enhancing the accuracy of document classification. In recent years, the fusion of deep learning techniques and text clustering has shown promise in extracting meaningful patterns and representations from textual data. This paper provides an in-depth exploration of various deep text clustering methodologies, assessing their efficacy in improving document classification accuracy. Delving into the core of deep text clustering, the paper investigates various feature representation techniques, ranging from conventional word embeddings to contextual embeddings furnished by BERT and GPT models.By critically reviewing and comparing these algorithms, we shed light on their strengths, limitations, and potential applications. Through this comprehensive study, we offer insights into the evolving landscape of document analysis and classification, driven by the power of deep text clustering algorithms.Through an original synthesis of existing literature, this research serves as a beacon for researchers and practitioners in harnessing the prowess of deep learning to enhance the accuracy of document classification endeavors.

Determination of scattering and Urbach absorption contributions to the light extinction in PTFE films by using graphical representation technique and numerical solution of the inverse problem

Ellipsometrically obtained spectral dependences of ordinary αxy and extra-ordinary αz extinction/attenuation coefficients within the spectral range λ = 300…980 nm of uniaxially anisotropic polytetrafluoroethylene (PTFE) films were analyzed. We considered the capabilities and specific features of the graphical representation technique for determining the contribution of Rayleigh scattering and Urbach absorption to light attenuation in the spectral range beyond fundamental absorption. It has been shown that the graphical approach enables to estimate these contributions qualitatively, semi-quantitatively or quantitatively, depending on the situation. The conclusions made using the analysis of graphical representation are confirmed by numerical solution of the inverse problem via simulation of the αxy (λ), αz (λ) experimental dependences within the framework of a best-fit procedure. Being based on both of these approaches, we have ascertained that, in the as-prepared PTFE films, the so-called anomalous light scattering (ALS) with the spectral dependence of scattering coefficient αs ≈ as λ–p (p &gt; 4) takes place. Transformation of scattering from ALS to the Rayleigh one with p ≈ 4 due to annealing is accompanied by an increase of Urbach (subband) absorption. Both of these factors cause narrowing the dynamic range of extinction coefficient values. Both scattering and absorption coefficients are higher for the component of light polarized along the normal to the substrate as compared to the component polarized in parallel to it. The relationship between observed behavior of the scattering and absorption coefficients and the film structure has been discussed.