Search

Dialogue-based multiple-choice machine reading comprehension (MRC) is one of most difficult and novel tasks because it requires more advanced reading comprehension skills, such as speaker’s intention analysis, non-extractive reasoning, commonsense knowledge. Previous models usually only compute attention scores from the fixed representation of entire dialogue, and also do not fully consider the contribution of dialogue, question, options, and their combinations respectively. In this paper, we introduce Utterance-focusing Multiway-matching Network (UMN), a simple but effective human mimicking model for dialogue-based multiple-choice MRC. First, two utterance-focusing mechanisms called ParaUF and AutoUF are proposed to extract the utterances that are most relevant to the question and option: ParaUF gets the bilinear weighted distance between each utterance of dialogue and question and option during training while AutoUF obtains the scores by the relevance, overlap and coverage (ROC) rules before training process. Second, we adopted the multiway-matching mechanism to capture the relationship among the question, option and selected utterances through calculating the attention weights between the quadruplet of four sequences: utterances, question, option and the concatenation of each two. We evaluate the proposed model on dialogue-based multiple-choice MRC tasks, DREAM, and outperformed recently published methods under the same pre-trained model. A series of detailed analysis is also conducted to interpret the differences of two utterance-focusing mechanisms and the effectiveness of the proposed multiway-matching mechanism.

A mathematic model considering the interaction of bubbles was developed to simulate bubbles behavior in a continuous slab casting mold. The flow of molten steel is calculated with an Eulerian model coupled with a Lagrangian model for bubbles. Modified bubble coalescence and breakup models are established as user-defined functions of FLUENT software. Collisions between bubbles are described by means of hard-sphere model. The break-up mold is implemented with a break-up constrain based on the critical Weber number augmented a model for daughter bubble size distribution. The daughter bubble fraction is calculated with a tanh function, which has a U-shape distribution with lower bounds. Small bubbles were produced by the breakup of large bubbles. The predicted size and number of captured bubbles obtained with the mathematic model agree well with the plant measurements. The size and number distributions of small bubbles near the advancing solidification shell were obtained with simulation results. The results show that the bubble size and number at both the narrow and wide faces decrease with the increase of distance from the meniscus, and they get close to a constant in the deep region. Meanwhile, the initial bubble size has no obvious effect on the bubble distribution in the mold at the operation conditions in the present work.

In this study, a nanoflower structured WO3 thin film has been prepared on an etched titanium substrate using a WO3 seeds-assisted hydrothermal reaction method. The WO3 nanoflower-like structure appeared when the growth time arrived at 4.75 h. Subsequently, this film went into a rapid growth stage. The nanoflower (NF) structured WO3 thin film with 8 h of hydrothermal reaction possessed the best photoelectrochemical performance. The photoinduced current density was 2.0 mA·cm−2 at a bias potential of 1.23 V (vs. Ag/AgCl) under the illumination by a simulated sunlight (100 mW/cm2) with a filter air mass (AM) 1.5 G. The improvement of the photoelectrochemical performance of the nanoflower structured WO3 can be attributed to three aspects. Firstly, the light absorption capacity could be increased by the nanoflower structure. Secondly, the reactive sites could be enhanced due to the increased specific surface area of the nanoflower structured WO3. Finally, a direct transfer channel for the photogenerated electrons could be formed by the definite 2-dimentional WO3 nanoflakes linked with the Ti substrate.

Copper element was employed to achieve the interlayer doping and intralayer doping in g-C3N4 (CN). Univalent Cu single atom prefer to dope into the interlayer site of CN to form a steady bridged bond structure. Further increasing the doping amount of Cu, except for the univalent Cu at the interlayer, some divalent Cu single atoms can be doped on the intralayer site of CN to form an unstable but active-catalysis structure. Electrochemical and theoretical calculation results reveal that Cu-interlayer doping can form an energy gradient between the different CN layers to decrease the charge transfer energy barrier between them. More importantly, these interlayer photogenerated electrons will converge on the divalent Cu, a catalysis active site for the photogenerated electrons, inducing more superoxide radicals and high activity. This two-sites modified CN, can mineralize a refractory antibiotic Norfloxacin (NOR) to CO2 and H2O completely under visible light stimulation, and the two-sites co-doping method provides a reference for CN photocatalyst modification to solve the questions of insufficient redox energy and low photons quantum efficiency of CN.

The hydrogen storage properties and mechanisms of the Ca(BH4)2-added 2LiNH2–MgH2 system were systematically investigated. The results showed that the addition of Ca(BH4)2 pronouncedly improved hydrogen storage properties of the 2LiNH2–MgH2 system. The onset temperature for dehydrogenation of the 2LiNH2–MgH2–0.3Ca(BH4)2 sample is only 80 °C, a ca. 40 °C decline with respect to the pristine sample. Further hydrogenation examination indicated that the dehydrogenated 2LiNH2–MgH2–0.1Ca(BH4)2 sample could absorb ca. 4.7 wt% of hydrogen at 160 °C and 100 atm while only 0.8 wt% of hydrogen was recharged into the dehydrogenated pristine sample under the same conditions. Structural analyses revealed that during ball milling, a metathesis reaction between Ca(BH4)2 and LiNH2 firstly occurred to convert to Ca(NH2)2 and LiBH4, and then, the newly developed LiBH4 reacted with LiNH2 to form Li4(BH4)(NH2)3. Upon heating, the in situ formed Ca(NH2)2 and Li4(BH4)(NH2)3 work together to significantly decrease the operating temperatures for hydrogen storage in the Ca(BH4)2-added 2LiNH2–MgH2 system.