Semantic similarity

About: Semantic similarity is a research topic. Over the lifetime, 14605 publications have been published within this topic receiving 364659 citations. The topic is also known as: semantic relatedness.

Generating Natural Language Adversarial Examples through Probability Weighted Word Saliency.

Abstract: We address the problem of adversarial attacks on text classification, which is rarely studied comparing to attacks on image classification. The challenge of this task is to generate adversarial examples that maintain lexical correctness, grammatical correctness and semantic similarity. Based on the synonyms substitution strategy, we introduce a new word replacement order determined by both the word saliency and the classification probability, and propose a greedy algorithm called probability weighted word saliency (PWWS) for text adversarial attack. Experiments on three popular datasets using convolutional as well as LSTM models show that PWWS reduces the classification accuracy to the most extent, and keeps a very low word substitution rate. A human evaluation study shows that our generated adversarial examples maintain the semantic similarity well and are hard for humans to perceive. Performing adversarial training using our perturbed datasets improves the robustness of the models. At last, our method also exhibits a good transferability on the generated adversarial examples.

Semantic and schematic similarities between database objects: a context-based approach

Abstract: In a multidatabase system, schematic conflicts between two objects are usually of interest only when the objects have some semantic similarity. We use the concept of semantic proximity, which is essentially an abstraction/mapping between the domains of the two objects associated with the context of comparison. An explicit though partial context representation is proposed and the specificity relationship between contexts is defined. The contexts are organized as a meet semi-lattice and associated operations like the greatest lower bound are defined. The context of comparison and the type of abstractions used to relate the two objects form the basis of a semantic taxonomy. At the semantic level, the intensional description of database objects provided by the context is expressed using description logics. The terms used to construct the contexts are obtained from {\em domain-specific ontologies}. Schema correspondences are used to store mappings from the semantic level to the data level and are associated with the respective contexts. Inferences about database content at the federation level are modeled as changes in the context and the associated schema correspondences. We try to reconcile the dual (schematic and semantic) perspectives by enumerating possible semantic similarities between objects having schema and data conflicts, and modeling schema correspondences as the projection of semantic proximity with respect to (wrt) context.

Using measures of semantic relatedness for word sense disambiguation

Abstract: This paper generalizes the Adapted Lesk Algorithm of Banerjee and Pedersen (2002) to a method of word sense disambiguation based on semantic relatedness This is possible since Lesk's original algorithm (1986) is based on gloss overlaps which can be viewed as a measure of semantic relatedness We evaluate a variety of measures of semantic relatedness when applied to word sense disambiguation by carrying out experiments using the English lexical sample data of SENSEVAL-2 We find that the gloss overlaps of Adapted Lesk and the semantic distance measure of Jiang and Conrath (1997) result in the highest accuracy

System and method for measuring image similarity based on semantic meaning

Abstract: A method includes deriving a plurality of semantic categories for representing important semantic cues in images, where each semantic category is modeled through a combination of perceptual features that define the semantics of that category and that discriminate that category from other categories; for each semantic category, forming a set of the perceptual features comprising required features and frequently occurring features; comparing an image to said semantic categories; and classifying said image as belonging to one of said semantic categories if all of the required features and at least one of the frequently occurring features for that semantic category are present in said image. A database contains image information, where the image information includes at least one of already classified images, network locations of already classified images and documents containing already classified images. The database is searched for images matching an input query, comprising, e.g., an image, text, or both.

Anchor-PROMPT: Using Non-Local Context for Semantic Matching.

Abstract: Researchers in the ontology-design field have developed the content for ontologies in many domain areas. Recently, ontologies have become increasingly common on the WorldWide Web where they provide semantics for annotations in Web pages. This distributed nature of ontology development has led to a large number of ontologies covering overlapping domains, which researchers now need to merge or align to one another. The processes of ontology alignment and merging are usually handled manually and often constitute a large and tedious portion of the sharing process. We have developed and implemented Anchor-PROMPT—an algorithm that finds semantically similar terms automatically. Anchor-PROMPT takes as input a set of anchors—pairs of related terms defined by the user or automatically identified by lexical matching. AnchorPROMPT treats an ontology as a graph with classes as nodes and slots as links. The algorithm analyzes the paths in the subgraph limited by the anchors and determines which classes frequently appear in similar positions on similar paths. These classes are likely to represent semantically similar concepts. Our experiments show that when we use Anchor-PROMPT with ontologies developed independently by different groups of researchers, 75% of its results are correct. 1 Ontology Merging and Anchor-PROMPT Researchers have pursued development of ontologies— explicit formal specifications of domains of discourse—on the premise that ontologies facilitate knowledge sharing and reuse (Musen 1992; Gruber 1993). Today, ontology development is moving from academic knowledgerepresentation projects to the world of e-commerce. Companies use ontologies to share information and to guide customers through their Web sites. The ontologies on the World-Wide Web range from large taxonomies categorizing Web sites (such as on Yahoo!) to categorizations of products for sale and their features (such as on Amazon.com). The WWW Consortium is developing the Resource Description Framework (Brickley and Guha 1999), a language for encoding semantic information on Web pages in machine-readable form. Such encoding makes it possible for electronic agents searching for information to share the common understanding of the semantics of the data represented on the Web. Many disciplines now develop standardized ontologies that domain experts can use to share and annotate information in their fields. Medicine, for example, has produced large, standardized, structured vocabularies such as SNOMED (Price and Spackman 2000) and the semantic network of the Unified Medical Language System (Humphreys and Lindberg 1993). With this widespread distributed use of ontologies, different parties inevitably develop ontologies with overlapping content. For example, both Yahoo! and the DMOZ Open Directory (Netscape 1999) categorize information available on the Web. The two resulting directories are similar, but also have many differences. Currently, there are extremely few theories or methods that facilitate or automate the process of reconciling disparate ontologies. Ontology management today is mostly a manual process. A domain expert who wants to determine a correlation between two ontologies must find all the concepts in the two source ontologies that are similar to one another, determine what the similarities are, and either change the source ontologies to remove the overlaps or record a mapping between the sources for future reference. This process is both labor-intensive and error-prone. The semi-automated approaches to ontology merging that do exist today (Section 2) such as PROMPT and Chimaera analyze only local context in ontology structure: given two similar classes, the algorithms consider classes and slots that are directly related to the classes in question. The algorithm that we present here, Anchor-PROMPT, uses a set of heuristics to analyze non-local context. The goal of Anchor-PROMPT is not to provide a complete solution to automated ontology merging but rather to augment existing methods, like PROMPT and Chimaera, by determining additional possible points of similarity between ontologies. Anchor-PROMPT takes as input a set of pairs of related terms—anchors—from the source ontologies. Either the user identifies the anchors manually or the system generates them automatically. From this set of previously identified anchors, Anchor-PROMPT produces a set of new pairs of semantically close terms. To do that, Anchor-PROMPT traverses the paths between the anchors in the corresponding ontologies. A path follows the links between classes defined by the hierarchical relations or by slots and their domains and ranges. Anchor-PROMPT then compares the terms along these paths to find similar terms. For example, suppose we identify two pairs of anchors: classes A and B and classes H and G (Figure 1). That is, a class A from one ontology is similar to a class B in the other ontology; and a class H from the first ontology is similar to a class G from the second one. Figure 1 shows one path from A to H in the first ontology and one path from B to G in the second ontology. We traverse the two paths in parallel, incrementing the similarity score between each two classes that we reach in the same step. For example, after traversing the paths in Figure 1, we increment the similarity score between the classes C and D and between the classes E and F. We repeat the process for all the existing paths that originate and terminate in the anchor points, cumulatively aggregating the similarity score. The central observation behind Anchor-PROMPT is that if two pairs of terms from the source ontologies are similar and there are paths connecting the terms, then the elements in those paths are often similar as well. Therefore, from a small set of previously identified related terms, AnchorPROMPT is able to suggest a large number of terms that are likely to be semantically similar as well. Figure 1. Traversing the paths between anchors. The rectangles represent classes and labeled edges represent slots that relate classes to one another. The left part of the figure represents classes and slots from one ontology; the right part represents classes and slots from the other. Solid arrows connect pairs of anchors; dashed arrows connect pairs of related terms.