SentenceTransformers库更新V2.0.0
1 引言
SentenceTransformers是一个用于句子、文本和图像嵌入的Python框架。使用这个框架可以计算多种语言的句子/文本嵌入。然后对这些嵌入进行比较，这对于语义上的文本相似性、语义搜索或转述挖掘都非常有用。该框架以PyTorch和Transformers为基础，并且提供了大量为各种任务调整的预训练模型。这个笔记建立在去年《Data Mining---岩土工程的数据挖掘》和《使用Transformers确定句子之间的相似度》的基础之上，因为原来的预训练模型不能再使用，所以更新了目前SentenceTransformers库包括预训练模型。目前的安装版本是V2.0.0，PyTorch的安装版本是版本V1.7.1。
2 预训练模型
尽管SentenceTransformers有许多预训练模型，但目前作为测试只使用了如下4个模型：
(1) paraphrase-mpnet-base-v2
(2) paraphrase-MiniLM-L6-v2
(3) distiluse-base-multilingual-cased
(4) distiluse-base-multilingual-cased-v2
按照SentenceTransformers的说法，(1)是最精确的模型，(2)是运行速度最快的模型，因此在本次测试中使用了这两个预训练模型。
3 测试结果
我们想查询"intact rock bridges", 首先使用geotech-words-flashtext-combination.py取出相似的段落，其工作机理参看《岩石边坡工程大数据系统设计》，然后使用SentenceTransformers模型进行相似性查询。
(A) 测试“diavik slope stability”， 其结果如下：
(1) Analysis of slope stability in strong, fractured rock at the Diavik Diamond Mine, NWT
(2) Rock Slope Stability Analysis by Using Integrated Approach
(3) LARGE SCALE SLOPE STABILITY IN OPEN PIT MINING
(4) Influence of discontinuity persistence on modelled slope behaviour ---The effect of non-persistent discontinuities and intact rock bridges on the stability of rock slopes has been emphasized within the literature
(5) Vivas (2014) used Slope Model to investigate the influence of intact rock bridges and a groundwater surface on the potential failure mechanism of a bi-planar failure mode based on the Diavik A418 pit geometry.
大型露天矿的边坡稳定性一直是岩石工程研究的课题，比如Diavik Diamond Mine，其中岩石的脆性断裂和原岩岩桥对边坡稳定性的影响是近年来研究的一个主要方向。Vivas(2014) 使用Slope Model分析了岩桥和地下水位对边坡破坏机理的影响。
(B) 测试“step-path modeling”， 其结果如下：
(1) STEPSIM4, simulates potential step-path dipping joints and direct shear through intact rock bridges.
(2) Step-path characterisation in rock---This paper presents a recently developed integrated digital imaging-numerical modeling approach to characterize step-path discontinuity geometries and intact rock bridges in the field, and to analyze their influence on rock slope stability.
(3) Characterization of step-path failure mechanisms- A combined field based-numerical modeling study==>Although this was followed by several probabilistic limit equilibrium developments to incorporate step-paths into rock slope design, comparatively little research has focused on the description of step-paths and intact fracture of rock bridges.
(4) Discretization routine in UDEC can model step-path formation by slipping along coplanar joints while by-passing intact rock bridges.
(5) Step-path failure surfaces and intact rock bridges have been recognized as an important component of rock slope failure since Jennings (1970) first introduced the concept of twodimensional and three-dimensional step-path geometries.
