Similarity Assessment Technique Research Articles

On the instability of further pre-training: Does a single sentence matter to BERT?

We observe a remarkable instability in BERT-like models: minimal changes in the internal representations of BERT, as induced by one-step further pre-training with even a single sentence, can noticeably change the behaviour of subsequently fine-tuned models. While the pre-trained models seem to be essentially the same, also by means of established similarity assessment techniques, the measurable tiny changes appear to substantially impact the models’ tuning path, leading to significantly different fine-tuned systems and affecting downstream performance. After testing a very large number of combinations, which we briefly summarize, the experiments reported in this short paper focus on an intermediate phase consisting of a single-step and single-sentence masked language modeling stage and its impact on a sentiment analysis task. We discuss a series of unexpected findings which leave some open questions over the nature and stability of further pre-training.

Material and textural features-based retrieval using feature similarity mapping for synthetic images

Content Based image retrieval (CBIR) is one of the prominent application of image processing where image retrieval from the database is made automatic based on the content in it. The materialist details of the images contains high degree of information and can serve as important toll for analysis. Different variant of content based image retrieval work by locating an object in the image and then retrieving images with similar object in it. The method works using different similarity assessment techniques. In this research work, we have developed a new material feature and HMMD space CBIR technique which works by transforming the domain to HMMD space and then finding the material and Texture Features along with other relevant features. These features forms he feature vector and are then deployed to assess the similarity of the two images which then generates a relevant and irrelevant image search. The system has higher accuracy of assessment as verified by experimentation for precision, productivity and precision agriculture.

Taxo-Semantics: Assessing similarity between multi-word expressions for extending e-catalogs

Taxonomies, also named directories, are utilized in e-catalogs to classify goods in a hierarchical manner with the help of concepts. If there is a need to create new concepts when modifying the taxonomy, the semantic similarity between the provided concepts has to be assessed properly. Existing semantic similarity assessment techniques lack in a comprehensive support for e-commerce, as those are not supporting multi-word expressions, multilingualism, the import/export to relational databases, and supervised user-involvement. This paper proposes Taxo-Semantics, a decision support system that is based on the progress in taxonomy matching to match each expression against various sources of background knowledge. The similarity assessment is based on providing three different matching strategies: a lexical-based strategy named Taxo-Semantics-Label, the strategy Taxo-Semantics-Bk, which is using different sources of background knowledge, and the strategy Taxo-Semantics-User that is providing user-involvement. The proposed system includes a translating service to analyze non-English concepts with the help of the WordNet lexicon, can parse taxonomies of relational databases, supports user-involvement to match single sequences with WordNet, and is capable to analyze each sequence as (sub)-taxonomy. The three proposed matching strategies significantly outperformed existing techniques. Taxo-Semantics-Label could improve the accuracy result by more than 7% as compared to state-of-the-art lexical techniques. Taxo-Semantics-Bk could improve the accuracy compared to structure-based techniques by more than 8%. And, Taxo-Semantics-User could additionally increase the accuracy by on average 23%.

Multiview Similarity Assessment Technique of UML Diagrams

Software system development has been beset by consistency problem since its infancy. This problem is still greater during the software design stage, and are greater still in UML diagrams. This is principally caused because of the existence of multiple views among UML diagrams. The main issue that arise when assessing the similarity of UML diagrams from multiple views is how consistency can be maintained across different views during the comparison. Previous works have failed to address this issue. They only compare one type of diagram in one stage and another type of diagram in another stage, this manner of similarity assessment lead to incorrect similarity scores, thereby creating inconsistencies in the information conveyed by the UML artifacts during software reuse. This paper presents Multiview similarity assessment technique, in which the similarity of UML diagrams is computed from structural and behavioral views of software systems.

Using EHRs for Heart Failure Therapy Recommendation Using Multidimensional Patient Similarity Analytics.

Electronic Health Records (EHRs) contain a wealth of information about an individual patient's diagnosis, treatment and health outcomes. This information can be leveraged effectively to identify patients who are similar to each for disease diagnosis and prognosis. In recent years, several machine learning methods have been proposed to assessing patient similarity, although the techniques have primarily focused on the use of patient diagnoses data from EHRs for the learning task. In this study, we develop a multidimensional patient similarity assessment technique that leverages multiple types of information from the EHR and predicts a medication plan for each new patient based on prior knowledge and data from similar patients. In our algorithm, patients have been clustered into different groups using a hierarchical clustering approach and subsequently have been assigned a medication plan based on the similarity index to the overall patient population. We evaluated the performance of our approach on a cohort of heart failure patients (N=1386) identified from EHR data at Mayo Clinic and achieved an AUC of 0.74. Our results suggest that it is feasible to harness population-based information from EHRs for an individual patient-specific assessment.

Similarity Estimation for Images

In this paper, a spatial domain image deinterlacing method is presented. The goal of this method is to achieve high-quality vertically upsampled images while spending reasonable computational time. We propose directional edge detection approach to recognize edge directions with their own orientations. Assumed and designed number of classified orientations is 11 and each direction has 2 samples. All 22 pixels are categorized into 5 areas, according to the distance from the center pixel. The weighted average approach is applied to restore the missing pixel in a certain window. The employed weights are evaluated by similarity assessment technique. Simulation results show that the proposed method is able to outperform conventional methods in terms of objective (PSNR, SSIM, and consumed time) and subjective metrics.

A similarity assessment technique for effective grouping of documents

Document clustering refers to the task of grouping similar documents and segregating dissimilar documents. It is very useful to find meaningful categories from a large corpus. In practice, the task to categorize a corpus is not so easy, since it generally contains huge documents and the document vectors are high dimensional. This paper introduces a hybrid document clustering technique by combining a new hierarchical and the traditional k-means clustering techniques. A distance function is proposed to find the distance between the hierarchical clusters. Initially the algorithm constructs some clusters by the hierarchical clustering technique using the new distance function. Then k-means algorithm is performed by using the centroids of the hierarchical clusters to group the documents that are not included in the hierarchical clusters. The major advantage of the proposed distance function is that it is able to find the nature of the corpora by varying a similarity threshold. Thus the proposed clustering technique does not require the number of clusters prior to executing the algorithm. In this way the initial random selection of k centroids for k-means algorithm is not needed for the proposed method. The experimental evaluation using Reuter, Ohsumed and various TREC data sets shows that the proposed method performs significantly better than several other document clustering techniques. F-measure and normalized mutual information are used to show that the proposed method is effectively grouping the text data sets.

An extended Q-gram algorithm for calculating the relevance factor of products in electronic marketplaces

Intelligent agents offer a number of advantages when used in electronic markets. In such environments, intelligent agents can represent users acting as buyers or sellers. On the buyer’s side, an intelligent agent can undertake the responsibility of finding and purchasing products that meet the owner’s needs. In this process, the agent should decide if a product, offered by a seller, is relevant to the owner’s preferences. We propose an algorithm for calculating the relevance factor of a product based on the product description, constraints defined by the buyer and the product’s quality of service characteristics, such as the delivery time or the seller trust level. The proposed algorithm is based on widely known similarity assessment techniques. However, we also propose a new similarity assessment scheme based on the Q-grams technique. We describe the proposed solution and evaluate our methodology. The results show that the algorithm is an efficient way for the relevance factor calculation and quality of service characteristics play an important role in the calculation process. Quality of service factor calculation provides an additional level of intelligence in the proposed methodology.

Integrated Multimodal Copy-Paste Checking

Although the use of advanced computing and communications technology has made learning a significantly richer experience for learners, it also introduces greater ease in committing academic cheating. There is a growing increase of internet usage in educational communities. Powerful search engines and media gateways gives an ease of access to wide spreading and different types of information sources, this in turn reduces the need for learners to perform diligent research or study. A learner can easily copy information, found on the internet, relevant to the task at hand. E-learning systems would then need to incorporate extended functionalities to enable multimodal copy-paste checking. Besides the mere detection of text copy-paste activities, there is a need to determine deeper similarity relations in text and various other media formats. Furthermore, the introduction of a facility to monitor and check the student's ways of searching, reading and writing can improve the learning process. Our previous paper [1] describes the architecture and design of an integrated Copy- Paste system aimed to provide a platform addressing these concerns. This paper extends the work and shows our pioneering explorations in extending student modeling capability of e-learning systems. We discuss the use of layered similarity assessment techniques to add value to conventional copy-paste detection systems.

Comparison of detrital zircon age distributions by kernel functional estimation

A method is presented for assessing and statistically testing the similarity of detrital zircon U–Pb age distributions frequently used in provenance analysis, correlation and tectonic reconstructions. The method accounts for intrinsic measurement uncertainties by constructing kernel functional estimates of each set of age data that compensate for different degrees of measurement error by the application of varying levels of smoothing. The dissimilarity between these estimates can be quantified and provides a meaningful comparison between age distributions. A Monte Carlo permutation algorithm is used to test for equality between age distributions by grouping two sets of age data and then resampling the age distributions. The techniques are demonstrated with both synthetic and real data. Synthetic data illustrate the behaviour of the algorithms with data containing varying age modes and measurement uncertainties. Real data from Tasmania and southeastern Australia illustrates both the broad correlation with previous similarity assessment techniques and the applicability of the described methods to provenance, regional correlation and tectonic reconstruction.