Capturing user interests across extensive watch histories is critical for short-form video recommendation, yet scaling sequence length is limited by two bottlenecks: the semantic sparsity of atomic Video IDs and the quadratic computational complexity of Transformers. Traditional orthogonal Video IDs fail to capture content relationships and demand large embedding tables, while the quadratic complexity of self-attention restricts the maximum sequence length under strict industrial latency and resource constraints. In this work, we present a production-deployed framework for modeling ultra-long user behavior sequences at a billion-user scale. We first address the representation bottleneck by adopting content-native Semantic IDs. By utilizing depth-truncated, coarse-grained Semantic IDs, we shrink the embedding table size from corpus cardinality. This compact representation naturally generalizes to cold-start content through shared semantic prefixes. Second, to overcome the sequence scaling barrier, we introduce a Global-Aware Compression Transformer that leverages non-parametric temporal folding and unified global query integration to effectively condense the sequence, alleviating both the memory and computational bottlenecks of standard self-attention. Offline profiling on our computing infrastructure demonstrates an order-of-magnitude reduction in peak memory footprint and a drastic decrease in computational overhead. This efficiency gain enables supporting longer sequence lengths at an affordable cost in production, yielding substantial online gains in satisfied user engagement and satisfied content consumption in large-scale online A/B tests.

In politics, neologisms are frequently invented for partisan objectives.For example, "undocumented workers" and "illegal aliens" refer to the same group of people (i.e., they have the same denotation), but they carry clearly different connotations.Examples like these have traditionally posed a challenge to referencebased semantic theories and led to increasing acceptance of alternative theories (e.g., Two-Factor Semantics) among philosophers and cognitive scientists.In NLP, however, popular pretrained models encode both denotation and connotation as one entangled representation.In this study, we propose an adversarial neural network that decomposes a pretrained representation as independent denotation and connotation representations.For intrinsic interpretability, we show that words with the same denotation but different connotations (e.g., "immigrants" vs. "aliens", "estate tax" vs. "death tax") move closer to each other in denotation space while moving further apart in connotation space.For extrinsic application, we train an information retrieval system with our disentangled representations and show that the denotation vectors improve the viewpoint diversity of document rankings.

Ongoing information on snow and its extent is critical for understanding global water and energy cycles. Passive microwave data have been widely used in snow cover mapping for its long-time observation capabilities under all-weather conditions. But assessment of different passive microwave (PM) snow cover area (SCA) mapping algorithms have been rarely reported, especially in China. In this study, the performance of seven well documented successfully applied PM SCA mapping algorithms were tested using in situ snow depth measurements over China. The results shown in this study would contribute to the ongoing effort to improve the performance and applicability of PM SCA algorithms.

In this study, based on in-situ measurements collected in the North China Plain, adjusted vegetation water indexes were introduced on the basis of traditional vegetation water indexes to weak soil and background influence and propose multiscale winter wheat moisture inversion model that is suitable in North China Plain. The main conclusions are as follows: 1) The adjusted water indexes have a good correlation with VWC; 2) For a fixed spatial resolution and indices with same form, the longer wavelength remote sensing data were used, the better reversion performance would be achieved; 3) The trend of spatial variability of multiscale wheat moisture is consistent, which indicates that the adjusted vegetation water indexes have certain applicability in the North China Plain.

Passive microwave remote sensing provide soil moisture product. The problem of validation of soil moisture products is that the true soil moisture of such a large pixel is difficult to estimate. This paper is a preliminary research on how to measure the true soil moisture of a 30m×30m quadrat. The results suggest that 20 measurement in such a quadrat should gave a reliable estimation of the soil moisture of the quadrat. This could be used as a thumb rule in designing the validation plan of remote sensing soil moisture product. And also could be used to estimate the soil moisture distribution within a space-born microwave radiometer footprint.

Passive microwave remote sensing has demonstrated the potential for capturing the high temporal variability of the near-surface soil moisture, however the use of these data is limited by the poor spatial resolution. We compared two different downscaling methods using soil evaporative efficiency derived from Moderate-resolution Imaging Spectroradiometer (MODIS) to disaggregate AMSR-2 soil moisture product. Both methods used information from MODIS to obtain the distributed soil moisture map at regional scales and the results showed reasonable agreement with ground-based soil moisture observations. For Merlin downscaling results, the correlation coefficient and RMSE with ground measured data are 0.74 and 3.21%, and for UCLA downscaling results, the correlation coefficient and RMSE with ground measured data are 0.76 and 3.81%, respectively.