python金融风控评分卡模型和数据分析微专业课（博主亲自录制视频）：http://dwz.date/b9vv
最近需要对英文进行分词处理，希望能够实现还原英文单词原型，比如 boys 变为 boy 等。
简介
发现一个不错的工具Porter stemmer，主页是http://tartarus.org/~martin/PorterStemmer/。它被实现为N多版本，C、Java、Perl等。
下面是它的简单介绍：
Stemming, in the parlance of searching and information retrieval, is the
operation of stripping the suffices from a word, leaving its stem.
Google, for instance, uses stemming to search for web pages containing
the words connected, connecting, connection and connections when
you ask for a web page that contains the word connect.
There are basically two ways to implement stemming. The first approach
is to create a big dictionary that maps words to their stems. The
advantage of this approach is that it works perfectly (insofar as the
stem of a word can be defined perfectly); the disadvantages
are the space required by the dictionary and the investment required to
maintain the dictionary as new words appear. The second approach is to
use a set of rules that extract stems from words. The advantages of this
approach are that the code is typically
small, and it can gracefully handle new words; the disadvantage is that
it occasionally makes mistakes. But, since stemming is imperfectly
defined, anyway, occasional mistakes are tolerable, and the rule-based
approach is the one that is generally chosen.
In 1979, Martin Porter developed a stemming algorithm that, with minor
modifications, is still in use today; it uses a set of rules to extract
stems from words, and though it makes some mistakes, most common words
seem to work out right. Porter describes his
algorithm and provides a reference implementation in C at http://tartarus.org/~martin/PorterStemmer/index.html;
以前也曾经尝试过这个算法，但是因为下面的原因就放弃了！
比如输入 "create" 和 "created" ，得到的结果是 "creat" 。这点让我大失所望！这根本就没有把单词还原为原来的样子啊？
这次没办法，还是需要实现这样的功能，Google了半天，就发现Lucene里面有英文分词模块，可惜太复杂了，不适合我的这种简单应用。后来才知道，其实lucene里用的也就是这种方法。
于是乎，硬着头皮看了下他的主页，在FQA里发现了下面这句话！恍然大悟。
The purpose of stemming is to bring variant forms of a word together, not to map a word onto its ‘paradigm’ form.
Porter stemmer
并不是要把单词变为规范的那种原来的样子，它只是把很多基于这个单词的变种变为某一种形式！换句话说，它不能保证还原到单词的原本，也就
是"created"不一定能还原到"create"，但却可以使"create" 和 "created" ，都得到"creat" ！
实例
比如我输入 "create" 和 "created" ，它解析得到 "creat"
那么，只需要在查询时也做同样的处理即可！比如查询 "create created"，在数据库里查的时候，都只需要检索"creat"即可！
附录
This page was completely revised Jan 2006. The earlier edition is here.
This is the ‘official’ home page for distribution of the Porter Stemming Algorithm, written and maintained by its author, Martin Porter.
The Porter stemming algorithm (or ‘Porter stemmer’) is a process for removing the commoner morphological and inflexional endings from words in English. Its main use is as part of a term normalisation process that is usually done when setting up Information Retrieval systems.
History
The original stemming algorithm paper was written in 1979 in the Computer Laboratory, Cambridge (England), as part of a larger IR project, and appeared as Chapter 6 of the final project report,C.J. van Rijsbergen, S.E. Robertson and M.F. Porter, 1980. New models in probabilistic information retrieval. London: British Library. (British Library Research and Development Report, no. 5587).With van Rijsbergen’s encouragement, it was also published in,M.F. Porter, 1980, An algorithm for suffix stripping, Program, 14(3) pp 130−137.And since then it has been reprinted inKaren Sparck Jones and Peter Willet, 1997, Readings in Information Retrieval, San Francisco: Morgan Kaufmann, ISBN 1-55860-454-4.The original stemmer was written in BCPL, a language once popular, but now defunct. For the first few years after 1980 it was distributed in its BCPL form, via the medium of punched paper tape. Versions in other languages soon began to appear, and by 1999 it was being widely used, quoted and adapted. Unfortunately there were numerous variations in functionality among these versions, and this web page was set up primarily to ‘put the record straight’ and establish a definitive version for distribution.
Encodings
The ANSI C version that heads the table below is exactly equivalent to the original BCPL version. The BCPL version did, however, differ in three minor points from the published algorithm and these are clearly marked in the downloadable ANSI C version. They are discussed further below.
This ANSI C version may be regarded as definitive, in that it now acts as a better definition of the algorithm than the original published paper.
Over the years, I have received many encoding from other workers, and they are also presented below. I have a reasonable confidence that all these versions are correctly encoded.
language
author
affiliation
received
notes
ANSI C thread safe
me
Daniel van Balen
Oct 1999
slightly faster?
The Official Web Guide
Sep 2001
Csharp .NET compliant
Univerity of Paisley, Scotland
Nov 2002
Csharp again!
Dec 2015
"more like standard
C# code" (Brad)
Brunel University
Apr 2003
Visual Basic
VB7; .NET compliant
University of Piraeus, Greece
Jan 2005
Richard Heyes
Feb 2005
University of Georgia
Oct 2005
Humboldt-Universitaet zu Berlin
May 2007
National Institute of Standards and
Technology, Gaithersburg, MD USA
Sep 2007
Mar 2013
bitbucket link
Zalán Bodó
Babes-Bolyai University
Oct 2015
Indian Institute of Technology, Delhi
Nov 2015
Dhaval Dave
Jun 2016
github link
P.O. Jonsson
Jul 2016
sourceforge link
Joey Takeda
Feb 2019
github link
All these encodings of the algorithm can be used free of charge for any purpose. Questions about the algorithms should be directed to their authors, and not to Martin Porter (except when he is the author).
To test the programs out, here is a
Points of difference from the published algorithm
There is an extra rule in Step 2,(m>0) logi → logSo archaeology is equated with archaeological etc.
The Step 2 rule(m>0) abli → ableis replaced by(m>0) bli → bleSo possibly is equated with possible etc.
The algorithm leaves alone strings of length 1 or 2. In any case a string of length 1 will be unchanged if passed through the algorithm, but strings of length 2 might lose a final s, so as goes to a and is to i.
These differences may have been present in the program from which the published algorithm derived. But at such a great distance from the original publication it is now difficult to say.
It must be emphasised that these differences are very small indeed compared to the variations that have been observed in other encodings of the algorithm.
Status
The Porter stemmer should be regarded as ‘frozen’, that is, strictly defined, and not amenable to further modification. As a stemmer, it is slightly inferior to the Snowball English or Porter2 stemmer, which derives from it, and which is subjected to occasional improvements. For practical work, therefore, the new Snowball stemmer is recommended. The Porter stemmer is appropriate to IR research work involving stemming where the experiments need to be exactly repeatable.
Common errors
Historically, the following shortcomings have been found in other encodings of the stemming algorithm.
The algorithm clearly explains that when a set of rules of the type(condition)S1 → S2are presented together, only one rule is applied, the one with the longest matching suffix S1 for the given word. This is true whether the rule succeeds or fails (i.e. whether or not S2 replaces S1). Despite this, the rules are sometimes simply applied in turn until either one of them succeeds or the list runs out.
This leads to small errors in various places, for example in the Step 4 rules(m>1)ement →
(m>1)ment →
(m>1)ent →to remove final ement, ment and ent.
Properly, argument stems to argument. The longest matching suffix is -ment. Then stem argu- has measure m equal to 1 and so -ment will not be removed. End of Step 4. But if the three rules are applied in turn, then for suffix -ent the stem argum- has measure m equal to 2, and -ent gets removed.
The more delicate rules are liable to misinterpretation. (Perhaps greater care was required in explaining them.) So((m>1) and (*s or *t))ionis taken to mean(m>1)(s or t)ionThe former means that taking off -ion leaves a stem with measure greater than 1 ending -s or -t; the latter means that taking off -sion or -tion leaves a stem of measure greater than 1. A similar confusion tends to arise in interpreting rule 5b, to reduce final double L to single L.
Occasionally cruder errors have been seen. For example the test for Y being consonant or vowel set up the wrong way round.
It is interesting that although the published paper explains how to do the tests on the strings S1 by a program switch on the last or last but one letter, many encodings fail to use this technique, making them much slower than they need be.
FAQs (frequently asked questions)
#1. What is the licensing arrangement for this software?
This question has become very popular recently (the period 2008−2009), despite the clear statment above that ‘‘all these encodings of the algorithm can be used free of charge for any purpose.’’ The problem I think is that intellectual property has become such a major issue that some more formal statement is expected. So to restate it:
The software is completely free for any purpose, unless notes at the head of the program text indicates otherwise (which is rare). In any case, the notes about licensing are never more restrictive than the BSD License.
In every case where the software is not written by me (Martin Porter), this licensing arrangement has been endorsed by the contributor, and it is therefore unnecessary to ask the contributor again to confirm it.
I have not asked any contributors (or their employers, if they have them) for proofs that they have the right to distribute their software in this way.
(For anyone taking software from the Snowball website, the position is similar but simpler. There, all the software is issued under the BSD License, and for contributions not written by Martin Porter and Richard Boulton, we have again not asked the authors, or the authors’ employers, for proofs that they have such distribution rights.)
#2. Why is the stemmer not producing proper words?
It is often taken to be a crude error that a stemming algorithm does not leave a real word after removing the stem. But the purpose of stemming is to bring variant forms of a word together, not to map a word onto its ‘paradigm’ form.
And connected with this,
#3. Why are there errors?
The question normally comes in the form, why should word X be stemmed to x1, when one would have expected it to be stemmed to x2? It is important to remember that the stemming algorithm cannot achieve perfection. On balance it will (or may) improve IR performance, but in individual cases it may sometimes make what are, or what seem to be, errors. Of course, this is a different matter from suggesting an additional rule that might be included in the stemmer to improve its performance.
python机器学习生物信息学系列课（博主录制）：http://dwz.date/b9vw
欢迎关注博主主页，学习python视频资源
