Smoothing LDA Model for Text Categorization

Abstract. Latent Dirichlet Allocation (LDA) is a document level language model. In general, LDA employ the symmetry Dirichlet distribution as prior of the topic-words’ distributions to implement model smoothing. In this paper, we propose a data-driven smoothing strategy in which probability mass is allocated from smoothing-data to latent variables by the intrinsic inference procedure of LDA. In such a way, the arbitrariness of choosing latent variables'priors for the multi-level graphical model is overcome. Following this data-driven strategy,two concrete methods, Laplacian smoothing and Jelinek-Mercer smoothing, are employed to LDA model. Evaluations on different text categorization collections show data-driven smoothing can significantly improve the performance in balanced and unbalanced corpora.

Hydrogen-rich gas production from biomass air and oxygen/steam gasification in a downdraft gasifier

Biomass gasification is an important method to obtain renewable hydrogen, However, this technology still stagnates in a laboratory scale because of its high-energy consumption. In order to get maximum hydrogen yield and decrease energy consumption, this study applies a self-heated downdraft gasifier as the reactor and uses char as the catalyst to study the characteristics of hydrogen production from biomass gasification. Air and oxygen/steam are utilized as the gasifying agents. The experimental results indicate that compared to biomass air gasification, biomass oxygen/steam gasification improves hydrogen yield depending on the volume of downdraft gasifier, and also nearly doubles the heating value of fuel gas. The maximum lower heating value of fuel gas reaches 11.11 MJ/ N m(3) for biomass oxygen/steam gasification. Over the ranges of operating conditions examined, the maximum hydrogen yield reaches 45.16 g H-2/kg biomass. For biomass oxygen/steam gasification, the content of H-2 and CO reaches 63.27-72.56%, while the content Of H2 and CO gets to 52.19-63.31% for biomass air gasification. The ratio of H-2/CO for biomass oxygen/steam gasification reaches 0.70-0.90, which is lower than that of biomass air gasification, 1.06-1.27. The experimental and comparison results prove that biomass oxygen/steam gasification in a downdraft gasifier is an effective, relatively low energy consumption technology for hydrogen-rich gas production.

富钇稀土-铝合金制取的研究Study on Preparation of Yttrium-Rich RE-Al Alloys

我国江西龙南稀土矿是目前世界上储量最大的富钇稀土矿、研制具有多种用途的钇（Y）-铝（Al）或富钇混合稀土（Ymm）-铝中间合金，对于开拓我国龙南稀土矿的应用领域扩大稀土合金出口具有重要意义。基于这一背景并针对目前氟化物体系制取Ymm-Al合金时存在着电解温度高，腐蚀现象严重，电效偏低等缺点，本文系统开展了在氯化物熔盐体系中电解制取Ymm-Al合金的研究工作。本工作由三部分组成：在第一部分工作中，开展了熔盐电解所需要基本原料-无水稀土氯化物制取的工艺研究。利用化学分析和结构分析手段，弄清了干法氯化过程中YmmCl_3水解的机理，提出了减弱水解的措施，即YmmCl_3先在850-900 ℃灼烧1.5 + 0.2hr，脱掉吸附水并将碱式碳酸盐转化为氧化物，增加稀土氧化物的比表面。通过条件试验得到最佳工艺条件为：采用NH_4 Cl:Ymm_2 O_3 = 14:1（摩尔比）的配料比，每次投入氯化装置的原料量为0.26 - 0.36 kg, 在400-450 ℃氯化反应激烈开始后迅速降温至400 ℃以下，待物料粘结现象消失后，再行升温氯化。出料及后期控制温在475 ± 25 ℃。经过3.8 ± 0.2hr氯化，可制得水不溶物小于1%并符合熔盐电解要求的YmmCl_3原料。此新工艺与原有干法工艺相比，流程短，装置简单，不需密闭抽真空，成本低，适于制取任何量的优质熔盐电解所需氯化稀土原料。在第二部分工作中，利用上述YmmCl_3原料，以液态铝为阴极，在氯化物体系中进行熔盐电解，通过试验得出在小型试验规模制取Ymm-Al合金的最隹工艺条件为：电解质组成（重量比）40%YmmCl_3-1%NaF-59%等摩尔的NaCl-KCl；电解温度为790 ± 5 ℃；阴极电流密为0.7 - 0.02A/cm~2；电解电量为333 ± 5库仑/克铝，制得钇铝合金中Ymm含量为10 ± 2%。添加1%的NaF可消除阴极表面生成枝状物，减少合金中夹渣和熔盐中沉渣。在电解工作中，将方差分析应用于试验数据处理，方差分析结果表明，各种试验因素对电效有明显影响，试验数据可靠，试验误差在允许范围以内。在第三部分工作中，利用线性扫描伏安法测定了在最隹电解工艺条件下Y~（3+）和Ymm在液态铝及钼电极上的析出电位。测定结果表明：Y~（3+）和Ymm~（3+）在液态铝阴极上的析出电位比在钼阴极上偏正0.2 ～ 0.8伏，氟离子的加入要比不加氟时析出电位不有同程度的负移，但考虑到氟离了具有消渣作用，加入少量氟比物添加剂对提高电效有利。

On-Chip All-Optical Passive 3.55 Gbit/s NRZ-to-PRZ Format Conversion Using a High-Q Silicon-Based Microring Resonator

We report the experimental result of all-optical passive 3.55 Gbit/s non-return-to-zero (NRZ) to pseudo-return-to-zero (PRZ) format conversion using a high-quality-factor (Q-factor) silicon-based microring resonator notch filter on chip. The silicon-based microring resonator has 23800 Q-factor and 22 dB extinction ratio (ER), and the PRZ signals has about 108 ps width and 4.98 dB ER.

Photoluminescence spectroscopy of sputtering Er-doped silicon-rich silicon nitride films

Er-doped silicon-rich silicon nitride (SRN) films were deposited on silicon substrate by an RF magnetron reaction sputtering system. After high temperature annealing, the films show intense photoluminescence in both the visible and infrared regions. Besides broad-band luminescence centered at 780 nm which originates from silicon nanocrystals, resolved peaks due to transitions from all high energy levels up to ~2H_(11/2) to the ground state of Er~(3+) are observed. Raman spectra and HRTEM measurements have been performed to investigate the structure of the films, and possible excitation processes are discussed.

Pressure behavior of Te isoelectronic centers in S-rich ZnS1-xTex alloy

The photoluminescence from ZnS1-xTex alloy with 0 < x < 0.3 was investigated under hydrostatic pressure up to 7 GPa. Two peaks were observed in the alloys with x < 0.01, which are related to excitons bound to isolated Te isoelectronic impurities (Te-1 centers) and Te pairs (Te-2 centers), respectively. Only the Te-2 related emissions were observed in the alloys with 0.01 < x < 0.03. The emissions in the alloys with 0.03 < x < 0.3 are attributed to the excitons bound to the Te-n (n greater than or equal to 3) cluster centers. The pressure coefficient of the Te-1 related peak is 89(4) meV/GPa, about 40% larger than that of the band gap of ZnS. On the other hand, the pressure coefficient of the Te-2 related emissions is only 52(4) meV/GPa, about 15% smaller than that of the ZnS band gap. A simple Koster-Slater model has been used to explain the different pressure behavior of the Te-1 and Te-2 centers. The pressure coefficient of the Te-3 centers is 62(2) meV/GPa. Then the pressure coefficients of the Te-n centers decrease rapidly with further increasing Te composition.