Search

Neural document ranking models perform impressively well due to superior language understanding gained from pre-training tasks. However, due to their complexity and large number of parameters, these (typically transformer-based) models are often non-interpretable in that ranking decisions can not be clearly attributed to specific parts of the input documents. In this paper we propose ranking models that are inherently interpretable by generating explanations as a by-product of the prediction decision. We introduce the Select-and-Rank paradigm for document ranking, where we first output an explanation as a selected subset of sentences in a document. Thereafter, we solely use the explanation or selection to make the prediction, making explanations first-class citizens in the ranking process. Technically, we treat sentence selection as a latent variable trained jointly with the ranker from the final output. To that end, we propose an end-to-end training technique for Select-and-Rank models utilizing reparameterizable subset sampling using the Gumbel-max trick. We conduct extensive experiments to demonstrate that our approach is competitive to state-of-the-art methods. Our approach is broadly applicable to numerous ranking tasks and furthers the goal of building models that are interpretable by design. Finally, we present real-world applications that benefit from our sentence selection method., Comment: Accepted to ACM TOIS

Parallel sentence pairs play a very important role in many natural language processing tasks, especially cross-lingual tasks such as machine translation. So far, many Asian language pairs lack bilingual parallel sentences. As collecting bilingual parallel data is very time-consuming and difficult, it is very important for many low-resource Asian language pairs. While existing methods have shown encouraging results, they rely on bilingual data seriously or have some drawbacks in an unsupervised situation. To address these issues, we propose a new unsupervised similarity calculation and dynamic selection metric to obtain parallel sentence pairs in an unsupervised situation. First, our method maps bilingual word embedding by postdoc adversarial training, which rotates the source space to match the target without parallel data. Then, we introduce a new cross-domain similarity adaption to obtain parallel sentence pairs. Experimental results on real-world datasets show that our model can obtain better accuracy and recall on mining parallel sentence pairs. We also show that the extracted bilingual sentence corpora can significantly improve the performance of neural machine translation.

This interdisciplinary research aims to examine the effects of photos used in Arabic news websites on both comprehension and recall among readers. The dual-coding theory was used as a theoretical framework. The research used an experimental design of two groups to test the effects of photos on comprehension and recall of the news content included in two news websites created specifically for the experiment. Two groups of 80 university students have participated voluntarily in the experiment. The results indicated that the use of photos in the news website has increased significantly the levels of both comprehension and recall among respondents.

Is it always beneficial to create a new relationship (have a new follower/friend) in a social network? This question can be formally stated as a property of the centrality measure that defines the importance of the actors of the network. Score monotonicity means that adding an arc increases the centrality score of the target of the arc; rank monotonicity means that adding an arc improves the importance of the target of the arc relatively to the remaining nodes. It is known that most centralities are both score and rank monotone on directed, strongly connected graphs. In this paper, we study the problem of score and rank monotonicity for classical centrality measures in the case of undirected networks: in this case, we require that score, or relative importance, improve at both endpoints of the new edge. We show that, surprisingly, the situation in the undirected case is very different, and in particular that closeness, harmonic centrality, betweenness, eigenvector centrality, Seeley's index, Katz's index, and PageRank are not rank monotone; betweenness and PageRank are not even score monotone. In other words, while it is always a good thing to get a new follower, it is not always beneficial to get a new friend., arXiv admin note: text overlap with arXiv:2202.01044

Searchable encryption (SE) is a powerful technology that enables keyword-based search over encrypted data becomes possible. However, most SE schemes focus on exact keyword search which can not tolerate misspellings and typos. Existing fuzzy keyword search schemes only support fuzzy search within a limited similarity threshold d, the storage cost will grow exponentially or the precision of search results will greatly decrease as $d$ increases. Moreover, the current fuzzy keyword ranked search schemes consider only the keyword weight, and disregard the influence of keyword morphology similarity on the ranking. In this paper, we propose a scalable fuzzy keyword ranked search scheme over encrypted data under hybrid clouds architecture. We use the edit distance to measure the similarity of keywords and design an edit distance algorithm over the encrypted database, in which our scheme achieves fuzzy keyword search for any similarity threshold d with constant storage size and accurate search results. Furthermore, we design a two-factor ranking function combining keyword weight with keyword morphology similarity, which is utilized to rank the search results and enhance system usability. Extensive experiments are performed to demonstrate the trade-off of efficiency and security of the proposed scheme.

This study aims at investigating procedures of semantic and linguistic extraction of keywords from metadata of documents indexed in the Institutional Repository Unesp. For that purpose, a web crawler was developed, and such tool collected 325.181 keywords from authors, in all fields of knowledge, from February, 28th, 2013 to November, 10th, 2021. The preparation of the collection, extraction and analysis environment used the Python programming language, composed of three program libraries: library requests, which allows manipulation of hyperlinks of webpages visited through web crawler; beautifulSoup library, used to extract HTML data through webpage analysis; and pandas library, which has an open code (free software) and stands for providing tools for high performance data manipulation and analysis. The final listing consisted of 273,485 keywords, which represents 15.9% of the listing initially collected. Results indicated that the most recurring problem was the duplication of keywords, with 51,696 duplicated keywords, representing indicators of inconsistencies in the search for documents. It is concluded that the refinement of keywords assigned by authors eliminates the incorporation of a set of symbols that do not represent the authors’ keywords with the same spelling, but with upper/lower case variations or lexical variations indexing different documents.

We examine how freemium platforms can design social referral programs to encourage growth and engagement without sacrificing revenue. On the one hand, social referral programs generate new referrals from users who would not have paid for the premium features. On the other hand, they also attract new referrals from users who would have paid but prefer to invite others, resulting in more referrals but fewer paying users. We use data from a large-scale randomized field experiment in an online dating platform to assess the effects of adding referrals programs to freemium platforms and changing the referral requirements on users’ behavior, namely, on their decisions to invite, pay, and engage with the platform. We find that introducing referral programs in freemium platforms can significantly contribute to increasing the number of referrals at the expense of revenue. Platforms can avoid the loss in revenue by reserving some premium features exclusively for paying users. We also find that increasing referral requirements in social referral programs can work as a double-edged sword. Increasing the referral threshold results in more referrals and higher total revenue. Yet these benefits appear to come at a cost. Users become less engaged, decreasing the value of the platform for all users. We explore two mechanisms that help to explain the differences in users’ social engagement. Finally, and contrary to prior findings, we find that the quality of the referrals is not affected by the referral requirements. We discuss the theoretical and practical implications of our research. This paper was accepted by Chris Forman, information systems. Funding: This work was funded by Fundac¸ão para a Ciência e a Tecnologia [UID/ ECO/00124/2019, UIDB/00124/2020, UIDP/00124/2020, and Social Sciences DataLab - PINFRA/22209/2016], POR Lisboa, and POR Norte [Social Sciences DataLab, PINFRA/22209/2016]. Supplemental Material: The online appendix and data are available at https://doi.org/10.1287/mnsc.2022.4301 .

Fake news detection attracts many researchers' attention due to the negative impacts on the society. Most existing fake news detection approaches mainly focus on semantic analysis of news' contents. However, the detection performance will dramatically decrease when the content of news is short. In this paper, we propose a novel $fake news detection multi-task learning (FDML)$ model based on the following observations: 1) some certain topics have higher percentages of fake news; and 2) some certain news authors have higher intentions to publish fake news. FDML model investigates the impact of topic labels for the fake news and introduce contextual information of news at the same time to boost the detection performance on the short fake news. Specifically, the FDML model consists of representation learning and multi-task learning parts to train the fake news detection task and the news topic classification task, simultaneously. As far as we know, this is the first fake news detection work that integrates the above two tasks. The experiment results show that the FDML model outperforms state-of-the-art methods on real-world fake news dataset.

So-called spatial web queries retrieve web content representing points of interest, such that the points of interest have descriptions that are relevant to query keywords and are located close to a query location. Two broad categories of such queries exist. The first encompasses queries that retrieve single spatial web objects that each satisfy the query arguments. Most proposals belong to this category. The second category, to which this paper's proposal belongs, encompasses queries that support exploratory user behavior and retrieve sets of objects that represent regions of space that may be of interest to the user. Specifically, the paper proposes a new type of query, the top-k spatial textual cluster retrieval ( $k$ -STC) query that returns the top-k clusters that (i) are located close to a query location, (ii) contain objects that are relevant with regard to given query keywords, and (iii) have an object density that exceeds a given threshold. To compute this query, we propose a DBSCAN-based approach and an OPTICS-based approach that rely on on-line density-based clustering and that exploit early stop conditions. Empirical studies on real data sets offer evidence that the paper's proposals can find good quality clusters and are capable of excellent performance.

Due to the expansion of data, search engines encounter different obstacles for retrieving better relevant content to user’s search queries. Consequently, various retrieval and ranking algorithms have been applied to satisfy the result’s relevancy according to user’s needs. Unfortunately, indexing and ranking processes face several challenges to achieve highly accurate results, since most of the existing indexes and ranking algorithms crawl documents and web pages based on limited number of criteria that satisfy user needs. So, this research studies and observes how search engines work and which factors contribute to higher rankings results. The research also proposes a Multi Criteria Indexing and Ranking Model (MCIR) based on weighted documents and pages which depend on one or more ranking factors, aiming at building a model that achieves high performance, better relevant pages, and the ability to index and rank both online/offline pages and documents. The MCIR model was applied on three different experiments to compare documents and pages results in terms of ranking scores, based on one or more criteria of user’s preferences. The results of applying MCIR model proved that final pages ranking results depending on multi-criteria are better than using only one criterion, and some criteria have great effect on ranking results than other criteria. It was also observed that the MCIR model achieved high performance on indexing and ranking dataset up to 100 gigabytes.

Aspect-based Opinion Mining is a form of fine-grained Sentiment Analysis and it models the semasiological relationship between aspect terms and context words in a sentence. The presence of a variety of context words has a significant impact on a sentence's sentiment polarity. As a result, while designing a model, it is necessary to consider the interaction of aspects and context words. Although existing approaches have taken into account an aspect’s position in a sentence, much of the research works have not explored the use of Sentiment Lexicons with the Deep Learning algorithms. In this paper, we propose a framework for an Attention-based position-aware Bidirectional Long Short-Term Memory network for Aspect-based Opinion Mining that incorporates a Sentiment Intensity Lexicon. The aspect word’s pre-trained vector is adjusted to be closer to semantically and sentimentally similar nearest neighbors and further away from sentimentally dissimilar neighbors. The proposed framework calculates aspect weights by concatenating the external knowledge in the form of lexicon sentiment intensity scores with word embeddings and position information. The framework experiments on the SemEval 2014 dataset. The results of the experiments illustrate that injecting external knowledge into the Bidirectional Long Short-Term Memory network can improve classification accuracy significantly.

In recent years, RDF has gained popularity as a format for the standardized publication and exchange of information in the Web of Data. In this paper we introduce RDFCSA, a data structure that is able to self-index an RDF dataset in small space and supports efficient querying. RDFCSA regards the triples of the RDF store as short circular strings and applies suffix sorting on those strings, so that triple-pattern queries reduce to prefix searching on the string set. The RDF store is then represented compactly using a Compressed Suffix Array (CSA), a proved technology in text indexing that efficiently supports prefix searches. Our experiments show that RDFCSA provides a compact RDF representation, using less than 60% of the space required by the raw data, and yields fast and consistent query times when answering triple-pattern queries (a few microseconds per result). We also support join queries, a key component of most SPARQL queries. RDFCSA is shown to provide an excellent space/time tradeoff, typically using much less space than alternatives that compete in time., Comment: This work has been submitted to a Journal for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible

The online environment dramatically expands the number of ways people can encounter news but there remain questions of whether these abundant opportunities facilitate news exposure diversity. This project examines key questions regarding how internet users arrive at news and what kinds of news they encounter. We account for a multiplicity of avenues to news online, some of which have never been analyzed: (1) direct access to news websites, (2) social networks, (3) news aggregators, (4) search engines, (5) webmail, and (6) hyperlinks in news. We examine the extent to which each avenue promotes news exposure and also exposes users to news sources that are left leaning, right leaning, and centrist. When combined with information on individual political leanings, we show the extent of dissimilar, centrist, or congenial exposure resulting from each avenue. We rely on web browsing history records from 636 social media users in the US paired with survey self-reports, a unique data set that allows us to examine both aggregate and individual-level exposure. Visits to news websites account for about 2 percent of the total number of visits to URLs and are unevenly distributed among users. The most widespread ways of accessing news are search engines and social media platforms (and hyperlinks within news sites once people arrive at news). The two former avenues also increase dissimilar news exposure, compared to accessing news directly, yet direct news access drives the highest proportion of centrist exposure.

Retrieval of mathematical information from scientific documents is one of the crucial tasks. Numerous Mathematical Information Retrieval (MIR) systems have been developed, which mainly focus on the improvement over the indexing and the searching mechanism, the poor results obtained for evaluation measures depict major limitations of such systems. These enhance the scope of improvement and new innovations through the inclusion of functionalities, which can resolve the challenges of MIR system. Further, to improve the performance of the MIR systems, this paper proposed a formula embedding and generalization approach with the context, in addition to innovative relevance measurement technique. In this approach, documents are preprocessed by the document preprocessor module and extracted the formulas in Presentation MathML format with their context. The formula embedding and generalization modules of the proposed approach formed the binary vectors where ‘1’ represents the presence, and ‘0’ represents the absence of a particular entity in a formula, and subsequently, the vectors of formulas with context are indexed by the indexer. The innovative relevance measurement technique of the proposed approach ranked those documents first, which are retrieved by both formula embedding and generalization modules as compared to the individual one. The proposed approach has been tested on the MathTagArticles of Wikipedia of NTCIR-12, and the obtained results verify the significance of the context of the formula and the dissimilarity factor in the retrieval of mathematical information.

Traditional searchable encryption schemes based on the Term Frequency-Inverse Document Frequency (TF-IDF) model adopt the presence of keywords to measure the relevance of documents to queries, which ignores the latent semantic meanings that are concealed in the context. Latent Dirichlet Allocation (LDA) topic model can be utilized for modeling the semantics among texts to achieve semantic-aware multi-keyword search. However, the LDA topic model treats queries and documents from the perspective of topics, and the keywords information is ignored. In this paper, we propose a privacy-preserving searchable encryption scheme based on the LDA topic model and the query likelihood model. We extract the feature keywords from the document using the LDA-based Information Gain (IG) and Topic Frequency-Inverse Topic Frequency (TF-ITF) model. With feature keyword extraction and the query likelihood model, our scheme can achieve a more accurate semantic-aware keyword search. A special index tree is used to enhance search efficiency. The secure inner product operation is utilized to implement the privacy-preserving ranked search. The experiments on real-world datasets demonstrate the effectiveness of our scheme.

Many recent approaches of passage retrieval are using dense embeddings generated from deep neural models, called "dense passage retrieval". The state-of-the-art end-to-end dense passage retrieval systems normally deploy a deep neural model followed by an approximate nearest neighbor (ANN) search module. The model generates embeddings of the corpus and queries, which are then indexed and searched by the high-performance ANN module. With the increasing data scale, the ANN module unavoidably becomes the bottleneck on efficiency. An alternative is the learned index, which achieves significantly high search efficiency by learning the data distribution and predicting the target data location. But most of the existing learned indexes are designed for low dimensional data, which are not suitable for dense passage retrieval with high-dimensional dense embeddings. In this paper, we propose LIDER, an efficient high-dimensional Learned Index for large-scale DEnse passage Retrieval. LIDER has a clustering-based hierarchical architecture formed by two layers of core models. As the basic unit of LIDER to index and search data, a core model includes an adapted recursive model index (RMI) and a dimension reduction component which consists of an extended SortingKeys-LSH (SK-LSH) and a key re-scaling module. The dimension reduction component reduces the high-dimensional dense embeddings into one-dimensional keys and sorts them in a specific order, which are then used by the RMI to make fast prediction. Experiments show that LIDER has a higher search speed with high retrieval quality comparing to the state-of-the-art ANN indexes on passage retrieval tasks, e.g., on large-scale data it achieves 1.2x search speed and significantly higher retrieval quality than the fastest baseline in our evaluation. Furthermore, LIDER has a better capability of speed-quality trade-off., Accepted by VLDB 2023

Are group variability judgments made in an on-line or memory-based fashion? In a first experiment addressing this question, subjects made judgments of a group's variability, a judgment intended to be on-line (liking), or a judgment intended to be memory-based (religiousness). Variability judgments were made more slowly than on-line judgments and at the same speed as memory-based judgments. Independently of this effect, in-group variability was judged more rapidly than out-group variability. In addition to replicating these results, a second experiment using a minimal group paradigm demonstrated that the amount of similarity information recalled predicted both the latency and extremity of variability judgments. We discuss the implications of our conclusion that variability judgments are predominantly memory-based for models of group variability judgments.

We define a procedure for canonicalising SPARQL 1.1 queries. Specifically, given two input queries that return the same solutions modulo variable names over any RDF graph (which we call congruent queries), the canonicalisation procedure aims to rewrite both input queries to a syntactically canonical query that likewise returns the same results modulo variable renaming. The use-cases for such canonicalisation include caching, optimisation, redundancy elimination, question answering, and more besides. To begin, we formally define the semantics of the SPARQL 1.1 language, including features often overlooked in the literature. We then propose a canonicalisation procedure based on mapping a SPARQL query to an RDF graph, applying algebraic rewritings, removing redundancy, and then using canonical labelling techniques to produce a canonical form. Unfortunately a full canonicalisation procedure for SPARQL 1.1 queries would be undecidable. We rather propose a procedure that we prove to be sound and complete for a decidable fragment of monotone queries under both set and bag semantics, and that is sound but incomplete in the case of the full SPARQL 1.1 query language. Although the worst case of the procedure is super-exponential, our experiments show that it is efficient for real-world queries, and that such difficult cases are rare.

Recent years have seen enormous gains in core IR tasks, including document and passage ranking. Datasets and leaderboards, and in particular the MS MARCO datasets, illustrate the dramatic improvements achieved by modern neural rankers. When compared with traditional test collections, the MS MARCO datasets employ substantially more queries with substantially fewer known relevant items per query. Given the sparsity of these relevance labels, the MS MARCO leaderboards track improvements with mean reciprocal rank (MRR). In essence, a relevant item is treated as the "right answer", with rankers scored on their ability to place this item high in the ranking. In working with these sparse labels, we have observed that the top items returned by a ranker often appear superior to judged relevant items. To test this observation, we employed crowdsourced workers to make preference judgments between the top item returned by a modern neural ranking stack and a judged relevant item. The results support our observation. If we imagine a perfect ranker under MRR, with a score of 1 on all queries, our preference judgments indicate that a searcher would prefer the top result from a modern neural ranking stack more frequently than the top result from the imaginary perfect ranker, making our neural ranker "better than perfect". To understand the implications for the leaderboard, we pooled the top document from available runs near the top of the passage ranking leaderboard for over 500 queries. We employed crowdsourced workers to make preference judgments over these pools and re-evaluated the runs. Our results support our concerns that current MS MARCO datasets may no longer be able to recognize genuine improvements in rankers. In future, if rankers are measured against a single "right answer", this answer should be the best answer or most preferred answer, and maintained with ongoing judgments.

This article describes the capability of online data storage which has been enhanced by the emergence of cloud datacenter development. Distributed Hash Table (DHT) based image retrieval system using locality sensitive hash (LSH) has provided an efficient way to set up distributed Content Based Image Retrieval (CBIR) frameworks. However, with the fixed LSH function adopted, LSH and other codebook-based distributed retrieval systems are facing the problem of flexibility, and also are difficult to satisfy the user's demand. In this article, LRFMIR is proposed to introduce semantic search into DHT based CBIR system. LRFMIR is established on a DHT based network, where a flexible result truncating strategy is employed to fuse provided results by using multiple features measurements. Experiments show that LRFMIR provides a higher accuracy and recall rate than single feature employed retrieval systems, and possesses good load balancing and query efficiency performance.

Due to the large French vocabulary, how quickly retrieve and accurately identify the required vocabulary is still a big challenge in French learning. In view of the above problems, we introduce a deep learning algorithm in this study to upgrade and optimize the retrieval system of French words and optimize the acquisition speed of speech words data and the recognition accuracy of speech words, so as to meet the needs of users for word retrieval. The results show that the two training methods of SGD synchronous update network and alternate update network parameters for fast retrieval and extraction of French speech vocabulary reduce from a maximum of 11.65% to 4.25% in the WER criterion, with a maximum reduction of 7.4%; the two training methods of SGD synchronous update network and alternate update network parameters for fast retrieval and extraction of French speech vocabulary reduce from a maximum of 13.52% to 4.4% in the SER criterion. The training methods of fast retrieval and extraction of the SGD synchronous update network and alternate update network parameters in French speech vocabulary reduced from the highest 582 ms to 351 ms in the response time criterion, with a maximum reduction of 8.84%; the maximum reduction of 39.7%. In French speech vocabulary, SGD synchronous updating network and alternating updating network parameter algorithm are used to quickly retrieve and extract French words. When the number of iterations reaches 120, the model fitting accuracy of the training set reaches 90.05%, while the model can reach 94.5% in the test set. The system has a stronger generalization ability and a higher speech vocabulary recognition rate to meet the practical requirements.

This paper presents EXMARaLDA, a system for the computer-assisted creation and analysis of spoken language corpora. The first part contains some general observations about technological and methodological requirements for doing corpus-based pragmatics. The second part explains the system’s architecture and gives an overview of its most important software components – a transcription editor, a corpus management tool and a corpus query tool. The last part presents some corpora which have been or are currently being compiled with the help of EXMARaLDA.

The establishment of English and French corpora has provided great convenience for language application and research. However, in the field of Japanese language research, the construction of Japanese corpus has been slow due to the limitation of Japanese markup format. This paper discusses the scientificity, rationality, feasibility, and construction of “microcorpus” and “specialized corpus” for the application of Japanese corpus in Japanese language teaching. Textual corpus is a corpus that contains all text but no media such as audio and pictures, and it is mostly used in teaching. The Japanese language corpus is a corpus created by analyzing the existing Japanese classroom teaching situation in response to the current reform of Japanese language teaching mode. It is important for teaching Japanese in colleges and universities because it can help students to learn independently and improve their learning initiative and motivation. The teaching-based text corpus is a small corpus for Japanese language teaching and learning, which is derived from the teaching process. The corpus can be created independently on an ordinary PC machine through software, and the collected corpus can be imported into the corpus and used, which provides another way to improve students’ interest and learning ability on their own. Based on the search software, this paper selects representative novels from the Aozora Library and creates a simple corpus to serve the teaching of Japanese interpretation, which improves the teaching effect. At the same time, this paper discusses the boundaries and difficulties of individual self-built corpora with a view to making breakthroughs and progress in future research and development.

With the growing availability and popularity of sentiment-rich resources like blogs and online reviews, new opportunities and challenges have emerged regarding the identification, extraction, and organization of sentiments from user-generated documents or sentences. Recently, many studies have exploited lexicon-based methods or supervised learning algorithms to separately conduct sentiment analysis tasks; however, the former approaches ignore contextual information of sentences and the latter ones do not take sentiment information embedded in sentiment words into consideration. To tackle these limitations, we propose a new model named Sentiment Convolutional Neural Network (SentiCNN) to analyze the sentiments of sentences with both contextual and sentiment information of sentiment words, in which, contextual information is captured from word embeddings and sentiment information is identified using existing lexicons. We incorporate a Highway Network into our model to adaptively combine sentiment and contextual information from sentences by strengthening the connection between features of both sentences and their sentiment words. Furthermore, we propose three lexicon-based attention mechanisms (LBAMs) for our SentiCNN model to find the most important indicators of sentiments and make predictions more effectively. Experiments over two well-known datasets indicate that sentiment words, the Highway Network, and LBAMs contribute to sentiment analysis.

With the advance of cloud computing technology, increasingly more documents are encrypted before being outsourced to the cloud for great convenience and economic savings. In this paper, we propose a fast and accurate searchable encryption (FASE) scheme that supports accurate top-k multi-keyword retrieval. We utilize a homomorphic orderpreserving encryption algorithm to encrypt the index and query vectors. The encryption method can not only ensure that the calculation of relevance score (SIi * T) is not exposed to the cloud server, but also protect the privacy of ranking operator. Compared to the traditional method, there are no dummy keywords added to the query vector and document vector, and the top-k search precision of the FASE scheme is 100%. To improve the search efficiency, a large number of irrelevant documents are effectively filtered by matching the document mark vector and query mark vector. Furthermore, according to the two-round ranking of the keyword matching degree and the relevance score, not only more accurate search result is returned, but the search efficiency is also further improved. The theoretical analysis and experimental results show that the FASE scheme can achieve fast and accurate multi-keyword ranking search.

An important challenge for news fact-checking is the effective dissemination of existing fact-checks. This in turn brings the need for reliable methods to detect previously fact-checked claims. In this paper, we focus on automatically finding existing fact-checks for claims made in social media posts (tweets). We conduct both classification and retrieval experiments, in monolingual (English only), multilingual (Spanish, Portuguese), and cross-lingual (Hindi-English) settings using multilingual transformer models such as XLM-RoBERTa and multilingual embeddings such as LaBSE and SBERT. We present promising results for "match" classification (86% average accuracy) in four language pairs. We also find that a BM25 baseline outperforms or is on par with state-of-the-art multilingual embedding models for the retrieval task during our monolingual experiments. We highlight and discuss NLP challenges while addressing this problem in different languages, and we introduce a novel curated dataset of fact-checks and corresponding tweets for future research., Accepted to De-Factify Workshop at AAAI 2022

Tourism industry is an important industry for many countries because this industry can bring a lot of profit and worker opportunity. Tourism enterprise is the element in the tourism industry. Many scholars execute research to understand the compatibility of tourism enterprise. However, past research evaluates the compatibility ability of tourism enterprise based on tourism enterprise entire resource, but past research does not evaluate the compatibility ability of tourism enterprise in detail under consumers’ viewpoint. The goal of the work is to build the framework to evaluate the competitive ability of tourism enterprises based on tourism resource in each tourism destination. Based on this framework, tourism enterprise can understand its competitive ability in special tourism regions. Some advantages of the proposed method and future research should be taken over finally.

Image captioning has emerged as an interesting research field in recent years due to its broad application scenarios. The traditional paradigm of image captioning relies on paired image-caption datasets to train the model in a supervised manner. However, creating such paired datasets for every target language is prohibitively expensive, which hinders the extensibility of captioning technology and deprives a large part of the world population of its benefit. In this work, we present a novel unpaired cross-lingual method to generate image captions without relying on any caption corpus in the source or the target language. Specifically, our method consists of two phases: (i) a cross-lingual auto-encoding process, which utilizing a sentence parallel (bitext) corpus to learn the mapping from the source to the target language in the scene graph encoding space and decode sentences in the target language, and (ii) a cross-modal unsupervised feature mapping, which seeks to map the encoded scene graph features from image modality to language modality. We verify the effectiveness of our proposed method on the Chinese image caption generation task. The comparisons against several existing methods demonstrate the effectiveness of our approach., Accepted by AAAI-2022

In this article, we examined the potential of the current version of WordNet and Google Translate API to enhance the quality of geodata source retrieval in the Dutch geoinformation portal (PDOK) using semantic keywords for the geographic phenomena requested. Keywords gathered from real users’ questions in natural language extracted in an English corpus. Then, these keywords were expanded using WordNet and Google Translate API. Lastly, the results of query expansion were evaluated compared to a manual gold standard and based on information retrieval metrics. Our study shows that the results of query expansion help users by reformulating good alternative queries.

Linked Open Datasets on the Web that are published as RDF can evolve over time. There is a need to be able to store such evolving RDF datasets, and query across their versions. Different storage strategies are available for managing such versioned datasets, each being efficient for specific types of versioned queries. In recent work, a hybrid storage strategy has been introduced that combines these different strategies to lead to more efficient query execution for all versioned query types at the cost of increased ingestion time. While this trade-off is beneficial in the context of Web querying, it suffers from exponential ingestion times in terms of the number of versions, which becomes problematic for RDF datasets with many versions. As such, there is a need for an improved storage strategy that scales better in terms of ingestion time for many versions. We have designed, implemented, and evaluated a change to the hybrid storage strategy where we make use of a bidirectional delta chain instead of the default unidirectional delta chain. In this article, we introduce a concrete architecture for this change, together with accompanying ingestion and querying algorithms. Experimental results from our implementation show that the ingestion time is significantly reduced. As an additional benefit, this change also leads to lower total storage size and even improved query execution performance in some cases. This work shows that modifying the structure of delta chains within the hybrid storage strategy can be highly beneficial for RDF archives. In future work, other modifications to this delta chain structure deserve to be investigated, to further improve the scalability of ingestion and querying of datasets with many versions.

Recently, the emergence of the digital language division and the availability of cross-lingual benchmarks make researches of cross-lingual texts more popular. However, the performance of existing methods based on mapping relation are not good enough, because sometimes the structures of language spaces are not isomorphic. Besides, polysemy makes the extraction of interaction features hard. For cross-lingual word embedding, a model named Cross-lingual Word Embedding Space Based on Pseudo Corpus (CWE-PC) is proposed to obtain cross-lingual and multilingual word embedding. For cross-lingual sentence pair interaction feature capture, a Cross-language Feature Capture Based on Similarity Matrix (CFC-SM) model is built to extract cross-lingual interaction features. ELMo pretrained model and multiple layer convolution are used to alleviate polysemy and extract interaction features. These models are evaluated on multiple language pairs and results show that they outperform the state-of-the-art cross-lingual word embedding methods.

With the rapid development of information collection technology and natural language processing technology, the construction of English-Chinese bilingual parallel corpus has developed rapidly. The scale of corpus and related technology have a large research space; and how to obtain effective data and knowledge from massive resources, in order to better serve the basic and applied research, is becoming a trend. Based on the research of parallel bilingual corpora at home and abroad, this article extracts the text features of medical English and constructs English-Chinese bilingual corpora with different levels aligned. Based on the statistical analysis of the distribution characteristics of phrase structures and the acquisition of characteristic knowledge, an English-Chinese bilingual translation model is constructed based on the bilingual corpus, and then the phrase structure knowledge in English and Chinese sentences is extracted from the data-driven English-Chinese bilingual corpus through the model. The results show that the accuracy of the test set is 91.63% and the F value is 90.05% under the condition of keeping the recall rate stable. The accuracy of the text structure features has been significantly improved, and the expected effect has been achieved through the test.

With the continuous advancement in technology, the tourism sector has developed to become one of the most important sectors globally in the modern period. These factors have given rise to the concept of “smart tourism,” which can be described as a step forward from conventional tourism. To achieve a true smart tourism experience, the appropriate services must be supplied to the correct user at the right time and in the most efficient and feasible manner. Keeping the importance of technology and tourism in consideration, the proposal and penetration of innovation-driven development strategies and smart tourism have become the focus of attention in the tourism industry at this stage. Smart tourism is comprised of a large number of tourists, devices, and operational processes which generate an enormous volume of tourism data. Handling such large amounts of tourism data in an effective and accurate manner is indeed an important thing to consider. To handle this issue, this study focuses on the construction and design of a smart tourism model based on big data technologies. This study explains in detail the relationship between smart tourism and big data and explores the construction of smart tourism applications in the context of big data. By explaining the relationship between smart tourism and big data, it is pointed out that the development of smart tourism needs to rely on the construction of a smart tourism application model under the background of big data. To this end, the role of smart tourism, building smart tourism platforms, improving information sharing mechanisms, exploring the implementation path of smart tourism application models, and further promoting the development of smart tourism, is of a great interest for the enterprises, scholars, and tourists. The proposed model is expected to be of a great help for the tourism industry.

Automatic Speech Recognition (ASR) is the use of computer hardware and software-based techniques to identify and process human voices. These systems used data from both male and female speakers. The majority of commercial ASR systems available on adult speech are working efficiently. Speech data collected from both male and female speakers were used in these systems. In recent decades, ASR systems for children have emerged, such as reading tutors, aids for foreign language learning, and computer games. Child ASR systems are essential but poorly understood in the field of computer speech recognition. The child data collection is a very complex task. Child corpus is not available publicly, and variability of children speakers and ASR developed for a particular age group is not suitable for other age groups. These are some of the reasons for less and ineffective child ASR systems. However, the non-availability of child corpus publicly is a primary reason for ineffective child ASR. Designing and developing child corpus is a very tedious task. Therefore, the primary focus of this state-of-the-art review is to discuss various challenges encountered while designing and developing child corpus.

POS (Parts of Speech) tagging, a vital step in diverse Natural Language Processing (NLP) tasks has not drawn much attention in case of Odia a computationally under-developed language. The proposed hybrid method suggests a robust POS tagger for Odia. Observing the rich morphology of the language and unavailability of sufficient annotated text corpus a combination of machine learning and linguistic rules is adopted in the building of the tagger. The tagger is trained on tagged text corpus from the domain of tourism and is capable of obtaining a perceptible improvement in the result. Also an appreciable performance is observed for news articles texts of varied domains. The performance of proposed algorithm experimenting on Odia language shows its manifestation in dominating over existing methods like rule based, hidden Markov model (HMM), maximum entropy (ME) and conditional random field (CRF).

Technical QA (2)Chatbot4QR can rapidly respond to the participants after receiving a query within ~1.3 seconds; (3)The refined queries contribute to retrieving more relevant SO questions than nine baseline approaches. For more than 70% of the participants who have preferred techniques on the query tasks, Chatbot4QR significantly outperforms the state-of-the-art word embedding-based retrieval approach with an improvement of at least 54.6% in terms of Pre@k and NDCG@k; and (4)For 48%-88% of the assigned query tasks, the participants obtain more desired results after interacting with Chatbot4QR than directly searching from Web search engines (e.g., the SO search engine and Google) using the original queries.

This article provides an overview of methodological and technical issues that arise in the collection, indexing and use of spoken learner corpora, i. e. corpora containing spoken utterances of learners of a target language. After an introductory discussion of the most important special features of this type of corpus that distinguish it from written language learner corpora and spoken corpora with L1 speakers, we will go into more detail on questions of corpus design. The main part of the paper is then an overview of the methodological and technical procedures of the individual steps of collecting, indexing, providing and using spoken learner corpora. The main aim of this overview is to highlight practices that can be considered best practices according to the current state of research. Finally, we outline the challenges that still exist for this type of corpus.

The confluence of high performance computing algorithms and large scale high-quality data has led to the availability of cutting edge tools in computational linguistics. However, these state-of-the-art tools are available only for the major languages of the world. The preparation of large scale high-quality corpora for low-resource language such as Urdu is a challenging task as it requires huge computational and human resources. In this paper, we build and analyze a large scale Urdu language Twitter corpus Anbar. For this purpose, we collect 106.9 million Urdu tweets posted by 1.69 million users during one year (September 2018-August 2019). Our corpus consists of tweets with a rich vocabulary of 3.8 million unique tokens along with 58K hashtags and 62K URLs. Moreover, it contains 75.9 million (71.0%) retweets and 847K geotagged tweets. Furthermore, we examine Anbar using a variety of metrics like temporal frequency of tweets, vocabulary size, geo-location, user characteristics, and entities distribution. To the best of our knowledge, this is the largest repository of Urdu language tweets for the NLP research community which can be used for Natural Language Understanding (NLU), social analytics, and fake news detection.

Deaf and Hard-of-Hearing (DHH) audiences have long complained about caption qualities for many online videos created by individual content creators on video-sharing platforms (e.g., YouTube). However, there lack explorations of practices, challenges, and perceptions of online video captions from the perspectives of both individual content creators and DHH audiences. In this work, we first explore DHH audiences' feedback on and reactions to YouTube video captions through interviews with 13 DHH individuals, and uncover DHH audiences' experiences, challenges, and perceptions on watching videos created by individual content creators (e.g., manually added caption tags could create additional confidence and trust in caption qualities for DHH audiences). We then discover individual content creators' practices, challenges, and perceptions on captioning their videos (e.g., back-captioning problems) by conducting a YouTube video analysis with 189 captioning-related YouTube videos, followed by a survey with 62 individual content creators. Overall, our findings provide an in-depth understanding of captions generated by individual content creators and bridge the knowledge gap mutually between content creators and DHH audiences on captions., Comment: 26 pages

There are a lot of research studies that look at "fake news" from an Arabic online source, but they don't look at what makes those fake news spread. The threat grows, and at some point, it gets out of hand. That's why this paper is trying to figure out how to predict the features that make Arabic online fake news spread. It's using Naive Bayes, Logistic Regression, and Random forest of Machine Learning to do this. Online news stories that were made up were used. They are found by using Term Frequency-Inverse Document Frequency (TF-IDF). The best partition for testing and validating the prediction was chosen at random and used in the analysis. So, all three machine learning classifications for predicting fake news in Arabic online were done. The results of the experiment show that Random Forest Classifier outperformed the other two algorithms. It had the best TF-IDF with an accuracy of 86 percent. Naive Bayes had an accuracy rate of 84%, and Logistic Regression had an accuracy rate of 85%, so they all did well. As such, the model shows that the features in TF-IDF are the most essential point about the content of an online Arabic fake news.

In distributed information retrieval systems, information in web should be ranked based on a combination of multiple features. Linear combination of ranks has been the dominant approach due to its simplicity and efficiency. Such a combination scheme in distributed infrastructure requires that ranks in resources or agents are comparable to each other. The main challenge is how to transform the raw rank values of different criteria appropriately to make them comparable before any combination. In this manuscript, we will demonstrate how to rank Web documents based on its resource-provided information stream and how to combine and incorporate several raking schemas in one time. The system was tested on the queries provided by a Text Retrieval Conference (TREC), and our experimental results showed that it is robust and efficient compared with similar platforms that used offline data resources.

It has been shown that the performance of neural machine translation (NMT) drops starkly in low-resource conditions. Thai-Lao is a typical low-resource language pair of tiny parallel corpus, leading to suboptimal NMT performance on it. However, Thai and Lao have considerable similarities in linguistic morphology and have bilingual lexicon which is relatively easy to obtain. To use this feature, we first build a bilingual similarity lexicon composed of pairs of similar words. Then we propose a novel NMT architecture to leverage the similarity between Thai and Lao. Specifically, besides the prevailing sentence encoder, we introduce an extra similarity lexicon encoder into the conventional encoder-decoder architecture, by which the semantic information carried by the similarity lexicon can be represented. We further provide a simple mechanism in the decoder to balance the information representations delivered from the input sentence and the similarity lexicon. Our approach can fully exploit linguistic similarity carried by the similarity lexicon to improve translation quality. Experimental results demonstrate that our approach achieves significant improvements over the state-of-the-art Transformer baseline system and previous similar works.

In recent years, social web users in Arabic countries have been resorting to the dialects as a written language in their social exchanges. Arabic dialects derive from modern standard Arabic (MSA) and differ significantly from one country to another and one region to another. The use of these dialects has led to an increase of interest in the specificities of such informal languages and their automatic processing within the NLP community. In this work, we deal with the Tunisian dialect (TD) in particular. We address the issue of the automatic Latin to Arabic transliteration of TD language productions on the social web and propose an approach that models the transliteration as a sequence labeling task. At a word level, several techniques, based on machine and deep learning, have been tested for this study, using real word messages extracted from social networks. We experiment and compare three transliteration models: A Conditional Random Fields-based model (CRF), a Bidirectional Long Short-Term Memory based model (BLSTM), and a BLSTM based model with CRF decoding (BLSTM-CRF). The obtained results show that BLSTM-CRF, leads to the best performance, reaching 96.78% of correctly transliterated words. We also evaluate the BLSTM-CRF transliteration approach in context on a set of random TD messages extracted from the social web. We obtained a total error rate of 2.7%. 25% of which are context errors.

Unmanned aerial vehicle (UAV) inspection is an indispensable part of power inspection. In the process of power inspection, the UAV needs to obtain an efficient and feasible path in the complex environment. To solve this problem, a Golden eagle optimizer with double learning strategies (GEO-DLS) is proposed. The double learning strategies consist of personal example learning and mirror reflection learning. The personal example learning can enhance the search ability of the Golden eagle optimizer (GEO) and reduce the possibility of the GEO falling into the local optimum. The mirror reflection learning can improve the optimization accuracy of the GEO and accelerate the convergence speed of the GEO. To verify the optimization performance of the algorithm, the proposed GEO-DLS and several other algorithms were tested under the CEC2013 test suite. At the same time, the proposed GEO-DLS and the GEO were analyzed for population diversity and exploration-exploitation ratio. Finally, the proposed GEO-DLS is applied to the UAV path planning to generate the initial path, and the cubic B-spline curve is used to smooth the path. These experimental results show that the GEO-DLS has a good performance. The code can be publicly available at: https://www.mathworks.com/matlabcentral/fileexchange/98799-golden-eagle-optimizer-with-double-learning-strategies

The impact of Extended Boolean retrieval (EBR) has drastic changes over the decades and despite their significant advantages compared to either ranked keyword or pure Boolean retrieval. In particular, EBR models produce meaningful rankings; their query model allows the representation of complex concepts in an and-or format; and they are scriptable, in that the score assigned to a document depends solely on the content of that document, unaffected by any collection statistics or other external factors. Security is the major concern to make the queries retrieval and the entire data mining process. However, EBR is much more computationally expensive than the alternatives. We consider the implementation of the p-norm approach to EBR, and demonstrate that ideas used in the max-score and wand exact optimization techniques for ranked keyword retrieval can be adapted to allow selective by pass of documents via a low-cost screening process for this and similar retrieval models. We also propose term independent bounds that are able to further reduce the number of score calculations for short, simple queries under the extended Boolean retrieval model. Overall saving from 50 to 80 percent of the evaluation cost on test queries drawn from biomedical search. (Stefan Pohl, VOL. 24, NO. 6, JUNE 2012) Keywords: extended, boolen, model, EBR

In the past years, significant efforts have been made for new theories and models of descriptors for Content-Based Image Retrieval systems and many effective descriptors, which use color and texture, have been established. This article presents the analysis and modifications of descriptors that use color and texture for the image retrieval task. To provide a complete detailed, and fair analysis, exposing weaknesses in descriptors and ideas to correct them. We evaluated descriptors that use color and texture, with image sets and metrics found in the literature. We compared classical descriptors that only use one low-level characteristic with descriptors that use color and texture. The analysis showed discrepancies between the model and the implementation of one of the descriptors, as well as the descriptors with the best performance, their main weaknesses, and complications when we trying to correct them. likewise, we present variants that improve the image retrieval in some cases.

The paper discusses the development of SPARQL query editor. This is an actual question because of the growth of Semantic Web data – the data presented in RDF/OWL formats. A comparative analysis of different types of editors and their main features is provided. In this paper, we propose a description of a SPARQL editor that combines three most useful features: intelligent completions, query visualization comparison of query results provided by different reasoners or without them. The editor provides SPARQL support as LSP service, this approach is considered a modern way to implement language support. This paper also presents the editor’s screenshots.

With the development of Internet technology, the phenomenon of information overload is becoming more and more obvious. It takes a lot of time for users to obtain the information they need. However, keyphrases that summarize document information highly are helpful for users to quickly obtain and understand documents. For academic resources, most existing studies extract keyphrases through the title and abstract of papers. We find that title information in references also contains author-assigned keyphrases. Therefore, this article uses reference information and applies two typical methods of unsupervised extraction methods (TF*IDF and TextRank), two representative traditional supervised learning algorithms (Na\"ive Bayes and Conditional Random Field) and a supervised deep learning model (BiLSTM-CRF), to analyze the specific performance of reference information on keyphrase extraction. It is expected to improve the quality of keyphrase recognition from the perspective of expanding the source text. The experimental results show that reference information can increase precision, recall, and F1 of automatic keyphrase extraction to a certain extent. This indicates the usefulness of reference information on keyphrase extraction of academic papers and provides a new idea for the following research on automatic keyphrase extraction.

README and replication code for: The effects of higher-stakes teacher evaluation on office disciplinary referrals

Linked data (LD) have the capability to open up and share materials, held in libraries, archives and museums (LAMs), in ways that are restricted by many existing metadata standards. Specifically, LD interlinking can be used to enrich data and to improve data discoverability on the Web through interlinking related resources across datasets and institutions. However, there is currently a notable lack of interlinking across leading LD projects in LAMs, impacting upon the discoverability of their materials. This research describes the Novel Authoritative Interlinking for Semantic Web Cataloguing in Libraries (NAISC-L) interlinking framework. Unlike existing interlinking frameworks, NAISC-L was designed specifically with the requirements of the LAM domain in mind. The framework was evaluated by Information Professionals (IPs), including librarians, archivists and metadata cataloguers, via three user-experiments including a think-aloud test, an online interlink creation test and a field test in a music archive. Across all experiments, participants achieved a high level of interlink accuracy, and usability measures indicated that IPs found NAISC-L to be useful and user-friendly. Overall, NAISC-L was shown to be an effective framework for engaging IPs in the process of LD interlinking, and for facilitating the creation of richer and more authoritative interlinks between LAM resources. NAISC-L supports the linking of related resource across datasets and institutions, thereby enabling richer and more varied search queries, and can thus be used to improve the discoverability of materials held in LAMs.