Search

Co-occurrence information between words is the basis of training word embeddings; besides, Chinese characters are composed of subcharacters, words made up by the same characters or subcharacters usually have similar semantics, but this internal substructure information is usually neglected in popular models. In this paper, we propose a novel method for learning Chinese word embeddings, which takes full use of external co-occurrence context information and internal substructure information. We represent each word as a bag of subcharacter n-grams, and our model learns the vector representation corresponding to the word and its subcharacter n-grams. The final word embeddings are represented as the sum of these two kinds of vector representation, which makes the learned word embeddings can take into account both the internal structure information and external co-occurrence information possible. The experiments show that our method outperforms state-of-the-art performance on benchmarks.

Heuristics serve as a powerful tool in classical planning. However, due to some incompatibilities between classical planning and hierarchical planning, heuristics from classical planning cannot be easily adapted to work in the hierarchical task network (HTN) setting. In order to improve HTN planning performance by the use of heuristics from classical planning, a new HTN planning named SHOP-h planning algorithm is established. Based on simple hierarchical ordered planner (SHOP), SHOP-h implemented with Python is called Pyhop-h. It can heuristically select the best decomposition method by using domain independent state-based heuristics. The experimental benchmark problem shows that the Pyhop-h outperforms the existed Pyhop in plan length and time. It can be concluded that Pyhop-h can leverage domain independent heuristics and other techniques both to reduce the domain engineering burden and to solve more and larger problems rapidly especially for problems with a deep hierarchy of tasks.

In this paper, a sequence-to-sequence based hybrid neural network model is proposed for abstractive summarization. Our method utilizes Bi-directional Long Short-Term Memory (Bi-LSTM) and multi-level dilated convolutions (MDC) to capture the global semantic information and semantic-unit level information, respectively. In decoding phrase, our model generates words according to summary relevant information captured by attention mechanism. Experiment shows that this proposed model outperforms several strong baselines on both of Gigawords corpus and DUC-2004 task.

Domain knowledge of hierarchical task network (HTN) usually involves logical expressions with predicates. One needs to master two different languages which are used to describe domain knowledge and implement planner. This has presented enormous challenges for most programmers who are not familiar with logical expressions. To solve the problem a method of state variable representation from the programmer’s point of view is introduced. This method has powerful expressivity and can unify the representations of domain knowledge and planning algorithm. In Pyhop a HTN planner written in Python, methods and operators are all as ordinary Python functions rather than using a special-purpose language. Pyhop uses a Python object that contains variable bindings and does not include a horn-clause inference engine for evaluating preconditions of operators and methods. By taking a simple travel-planning problem, it shows that the method is easy to understand and integrate planning into ordinary programming.