Article Per Year (5 Year)
p-Index From 2021 - 2026
0.562
P-Index
This Author published in this journals
All Journal Journal of Technology Informatics and Engineering
Yinchen Shi
Computer Science, New York University, NY, USA
Computer Science & IT

Published : 3 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 3 Documents
Search

 1 
IntentRouter-QAC: Distilling LLM-Derived Intent Signals into Small Language Models for Context-Aware Autosuggest and Search Entry Routing Haowei Tu; Yinchen Shi; Sophia Chen
Journal of Technology Informatics and Engineering Vol. 5 No. 1 (2026): APRIL | JTIE : Journal of Technology Informatics and Engineering
Publisher : University of Science and Computer Technology

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.51903/jtie.v5i1.563

Abstract

Search boxes increasingly function as routing interfaces: a partial query may require a completion, immediate submission, a context-sensitive suggestion, or a fallback action. This paper evaluates IntentRouter-QAC as a lightweight lexical routing framework and tests product retrieval independently. Five mutually exclusive operational outcomes are derived within the 20,000-row AmazonQAC test file and evaluated with a strict session-disjoint temporal split; product-query provenance is not used as a route class. Across five seeds, the combined word/character TF-IDF router reached 0.473 mean accuracy (SD 0.031) and 0.330 mean macro-F1 (SD 0.018), indicating uneven performance across routes. On the 951-row temporal test set, context-nearest completion achieved 0.033 Success@10 and 0.025 MRR@10, compared with 0.029 and 0.023 for confidence-gated routing. The paired MRR@10 difference was 0.002 (95% CI -0.007 to 0.010; Holm-adjusted p = 1.000). Product retrieval on 480 WANDS queries and 42,994 products produced a different pattern: a word/character lexical hybrid reached 0.690 nDCG@10 and 0.504 Exact-MRR@10, significantly exceeding the title-only baseline on both measures. Route-aware decision-making is therefore useful as an architectural separation of entry actions, but it does not automatically improve exact autocomplete ranking over a strong session-context baseline.
Token-Burst-Aware Capacity Planning for LLM Inference Services: Request Arrival, Token Demand, and Failure Risk Modeling from BurstGPT Traces Jiayi Nie; Yinchen Shi; Lucas Zhao
Journal of Technology Informatics and Engineering Vol. 5 No. 2 (2026): AUGUST | JTIE : Journal of Technology Informatics and Engineering
Publisher : University of Science and Computer Technology

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.51903/jtie.v5i2.565

Abstract

Large language model (LLM) inference services convert request arrivals into coupled input-token prefill and output-token decoding workloads. Capacity planning therefore depends on token volume, temporal bursts, service mix, queueing behavior, and reliability signals rather than request counts alone. This study evaluates an integrated planning pipeline on BurstGPT v1.1, comprising 5,288,173 raw requests over 121 trace days and 5,188,507 completed requests. Strictly chronological experiments aggregate demand, forecast hourly completed tokens, detect minute-level burst pressure, estimate zero-response risk, simulate capacity policies, and replay representative test hours in Vidur. Random forest, selected on the validation interval, achieved 64.73% weighted absolute percentage error (WAPE) on the locked test interval; XGBoost achieved the lowest test WAPE (64.57%), while the last-hour baseline reached 67.42%, indicating limited forecastability under a pronounced level shift. A seasonal-residual burst detector achieved F1 = 0.653, although burst prevalence was sensitive to rolling-horizon and quantile settings. For minute-level zero-response risk, raw XGBoost achieved ROC-AUC = 0.813 and average precision = 0.116; isotonic calibration improved the Brier score (0.0230–0.0204) and 10-bin expected calibration error (0.0262–0.0122), despite low F1. Static P90/P95 capacity eliminated under-provisioned test hours at cost indices of 13.10 and 18.09. More economical dynamic baselines achieved 12.65% under-provisioned hours at a cost index of 2.37 and 13.63% at 2.43. The validation-selected random-forest policy was cheaper but less reliable (34.31% at 1.23). Vidur replay linked normalized demand to A100/H100 GPU counts, latency, batching, and memory pressure. The results support conservative interpretation of point forecasts and validation of reserve rules under distribution shift, rare-event calibration, and serving-stack constraints.
Psychology-Informed Live-Commerce Analytics: Structural Modeling, Predictive Validation, and Governed Script Scoring for TikTok Shopping Xiaochen Li; Yinchen Shi; Jason Zhang
Journal of Technology Informatics and Engineering Vol. 5 No. 1 (2026): APRIL | JTIE : Journal of Technology Informatics and Engineering
Publisher : University of Science and Computer Technology

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.51903/jtie.v5i1.568

Abstract

This paper examines the psychological factors associated with impulsive-purchase propensity in TikTok Live Commerce and translates those associations into an auditable analytics workflow. It presents a secondary analysis of a public survey workbook containing 415 Indonesian responses on Social Attraction, fear of missing out, Narrative Involvement, Telepresence, Parasocial Interaction, Social Presence, and Impulsive Purchase. The analysis combines measurement diagnostics, a bootstrapped composite structural model, repeated cross-validated classification and continuous prediction, calibration assessment, a coefficient-based illustrative scenario analysis, and external validation of an ethical microcopy screen. In the full sample, Narrative Involvement has the largest association with Parasocial Interaction (β = .516) and Social Presence (β = .496); Parasocial Interaction and Social Presence jointly account for 67.5% of the in-sample variance in Impulsive Purchase. The pattern remains in the quality-screened TikTok-viewer subset (n = 238; Impulsive Purchase R² = .647). For the quality-screened subset, the strongest full-feature classifier is Random Forest (ROC-AUC = .938 ± .031 across 50 held-out folds), while Ridge regression gives the largest mean continuous-outcome R² (.622 ± .114). On 2,356 ec-darkpattern strings, a page-grouped hybrid word/character classifier reaches F1 = .959 ± .011 and ROC-AUC = .989 ± .003. The scenario scores are deterministic implications of the estimated coefficients rather than observed purchase effects. The technical contribution is a governed decision-support architecture that keeps structural association, contemporaneous prediction, scenario prioritization, ethical screening, human review, and prospective testing distinct.
Co-Authors Haowei Tu Jason Zhang Jiayi Nie Lucas Zhao Sophia Chen Xiaochen Li