Journal of Information Science and Engineering
- OpenAccess
社團法人中華民國計算語言學學會,正常發行
選擇卷期
- 期刊
- OpenAccess
A stacked recurrent neural network (sRNN) with gated recurrent units (GRUs) and jointly optimized soft time-frequency mask was proposed for extracting target musical instrument sounds from a mixture of instrumental sound. The sRNN model stacks and links multiple simple recurrent neural networks (RNNs), which makes sRNN an excellent model with temporal dynamic behavior and real deepness. The GRU improves the gate foundations of long short-term memory and reduces the operating time. Experiments were conducted to test the proposed method. A musical dataset collected from real instrumental music was used for training and testing; electric guitar and drum sounds were the target sounds. Objective and subjective assessment scores obtained for the proposed method were compared with those obtained for two models, namely Wave-U-Net and SH-4stack, and a conventional RNN model. The results indicated that electric guitar and drum sounds can be successfully extracted through the proposed method.
- 期刊
- OpenAccess
In this paper, we introduce a machine learning technique to estimate the vector encoded by a Variational Autoencoder (VAE) model, without the need of explicitly sampling the vector from the VAE's latent space. The feasibility of our approach is evaluated in the field of music interpolation composition, by means of the Hsinchu Interpolation MIDI Dataset that was created. A novel dual architecture of VAE plus an additional neural network (VAE+NN) is proposed to generate a polyphonic harmonic bridge between two given songs, smoothly changing the pitches and dynamics of the interpolation. The interpolations generated by the VAE+NN model surpass a Random data baseline, a bidirectional LSTM model and the state-of-the-art interpolation approach in automatic music composition (VAE model with linear sampling of the latent space), in terms of reconstruction MSE loss. Furthermore, a subjective evaluation was done in order to ensure the validity of the metric-based results.
- 期刊
- OpenAccess
A significant area of Social Computing is the Social Network Service (SNS), known as Social Network Applications. In recent years, researchers have commonly used SNS as an instrument for linking and communicating. Homestay travel has been prevalent for a long time with the rise of social media. The goal of this study is to examine the influence of Social Network Service functions on homestay travel intention in Vietnam. Fourteen Social Network Service functions were summarized from the literature review and used as the variables influencing the purpose of homestay travel to develop a five-point Likert scale questionnaire for convenience sampling to perform an online survey. For further study, two hundred and twenty valid respondents were included. The GM(1, N) analysis showed interest sharing, photo sharing, and video sharing as the first to third most principal factors in their highly seasoned weighting towards homestay travel intention. On the other side, helping decision, helping interaction, and helping planning as the last three lists of weighting scores. Besides, multiple regression analysis shows that offering recommendations, helping planning, and sharing interest simultaneously predict homestay travel intention while the others don't. That means consumers would heavily rely upon the functions of sharing interest of social network services to evaluate their traveling options. It is proposed that homestay traveling vendors should pay more attention to the marketing of previous travelers' experiential interests to provoke the awareness of customers.
- 期刊
- OpenAccess
In the last decade, recommendation systems have gradually become the most important service for online business, which serve as sales assistants for e-commerce business increasing their profits. However, the conventional recommendation systems are usually confronted with two challenges. First, in online shopping contexts, users often browse products that they do not go on to order. The majority of action sequences are browsing-browsing rather than browsing-order. As a result, user actions are not a direct reflection of user preferences. Second, the popularity of sold products creates a skewed distribution that results in the problem of cold-start product for recommendation. In this paper, we present our research on developing a two-stage framework of hybrid recommendation system to tackle these two challenges for tourism product recommendation. In order to extract knowledge from users' implicit feedback, we develop the neighborhood structure of users and products in the multi-behavior interaction network that simultaneously incorporates the browsing and order behaviors. To ensure the coverage of cold products, we considered the metadata associated with products and extracted more features from the textual content to form a product-content knowledge graph. By embedding the multi-behavior network and product-content knowledge graph within the recommendation system, we were able to capture user preferences from implicit feedback and the relationships among products. To evaluate the proposed model, we conducted experiments on a real-world dataset. Experimental results indicate that the proposed approach outperforms several widely-used recommendation systems.
- 期刊
- OpenAccess
In this paper, we proposed a method to generate two different types of reading comprehension questions complying with types of question for language learning tests with the Transformer model and the seq2seq method. In recent years, many approaches have showed good results by treating question generation as a seq2seq task. These approaches were implemented with a question-answering (QA) dataset; however, few studies have considered a reading comprehension-based dataset. Therefore, this paper proposed a method to generate questions appropriate for reading comprehension tests from articles. Moreover, analysis of reading comprehension test questions revealed two primary types of the question's asking style: the commonly-used question (CM question) and the directly- related question (DR question). The characteristic of the two question types was different and therefore needs to design the generation models complying with the type of questions. We proposed a method to separate the two question types in the dataset and used two models to generate both types, comparing the result with the method that generates the two types of questions together. The positive rate for the proposed method's CM questions was 88% and for its DR questions was 49%, compared to 33% and 24%, respectively, for the comparative method. The evaluation showed that the proposed method could generate the two types of reading comprehension questions more effectively, with a positive rate increased by an average of 40%.
- 期刊
- OpenAccess
The artificial intelligence (AI) in Shogi has made rapid progress recently, owing to the recent establishment of a method of learning evaluation via self-play. In this paper, we applied this method to Mini-Shogi to verify the effect. Specifically, we used YaneuraOu Shogi engine to develop the Mini-Shogi program and trained a neural network-based evaluation function. Our program won all competitions in which we participated in 2020. Moreover, the experimental results suggest the second-move (White) advantage in Mini-Shogi.
- 期刊
- OpenAccess
This paper proposes a lyrics retrieval method based on word-embedding technology to perform music recommendations that matched a tourist attraction. In the proposed method, word importance is calculated by Term Frequency - Inverse Document Frequency (TFIDF) and Smooth Inverse Frequency (SIF). We built a vectorization model from the lyrics corpus using fastText with Continuous Bag of Words (CBOW) and applied this model to both the lyrics corpus and the tourist attraction reviews corpus to create word-embedding vectors for lyrics and tourist attraction reviews. And, the review vectors are integrated for each tourist attraction to generate tourist attraction vectors. Based on the similarity calculation between the tourist attraction vectors and the lyric vectors, the song with the most suitable lyrics for the tourist attraction comes to be the recommended result. Subjective evaluation experiments on the recommendation results of the proposed method was conducted. The results of the experiments showed that the subjects reacted positively to the lyrics of the recommended songs. However, their evaluations of the music of the recommended songs were not as positive as those accorded to the lyrics were. A joint study of acoustic information and text information is an approach to the issue for the future.
- 期刊
- OpenAccess
Access to information in multisource environments is facing many problems. One of them is the source selection problem. As more and more sources become available on the internet, how to select the relevant sources that meet the user needs is a big challenge. In this paper, we propose a multi-dimensional source selection approach based on topic modelling, which integrates both the social dimension and the intelligent dimension in order to optimize the source selection according to different user interests. Social tagging data is analyzed to discover relevant topics of user interests and latent relationships between users and sources based on topic modelling. By intelligently exploring a large search space of possible solutions, an (optimal) selection of sources is found using an intelligent method (a genetic algorithm). The proposed approach is evaluated on real data sources. The experimental results demonstrate that the proposed approach outperforms state-of-the-art source selection algorithms.
- 期刊
- OpenAccess
In multi-objective optimization problems, the objective space of fitness functions has a close relationship with the solution space. Extracting the optimal direction and optimal parameter information are very useful for the optimization process. This paper proposes multi-objective differential evolution algorithm with a clustering based objective space division and parameter adaptation (MODECD). L∞ metric matrix based optimal strategy is used to split the objective space into sub-spaces and to extract the optimal directions. A fitness value based parameter adaptation and mutation strategy are used to extract the optimal strategy information. The results with 20 benchmark tests show the competitiveness of the MODECD algorithm in both convergence speed and diversity of solution approximating the Pareto front. In addition, MODECD is used to optimize the fermentation process of sodium gluconate as an example of its superior performance in solving real-world problems.
- 期刊
- OpenAccess
Severity prediction on software bug reports is an important research issue. Recently, many studies have been conducted. Although previous studies have explored different features to facilitate bug severity assessment, the effectiveness of jointly considering these features is not investigated. In the work, multiple features of three facets are collected are studied. Moreover, this study employs a weight adjustment approach using particle swarm optimization (PSO) to find the most appropriate weights of these features. In the prediction framework, three classification models are used to study the influences of these features. The experimental results show that PSO-optimized multi-facet features with the Random Forests model can achieve the best average prediction performance.