Information Extraction - Science topic

What kind of disaster of English news reports do you want to?

I have Japanese one.

DEAR Matiam Essa

This text mining technique focuses on identifying the extraction of entities, attributes, and their relationships from semi-structured or unstructured texts. Whatever information is extracted is then stored in a database for future access and retrieval.The famous technique are:

With the increasing amount of text data, effective techniques need to be employed to examine the data and to extract relevant information from it. We have understood that various text mining techniques are used to decipher the interesting information efficiently from multiple sources of textual data and continually used to improve text mining process.

Nice Dear Muhammad Ali

Thanks Gopi Battineni , I will check the paper.

Very interesting regarding the knowledge of ITES - Following .

Hi,

In relation to the text cleaning part of your question, we can make a list, containing your example and Arturo's:

1. Line feeds ("\n") could be invisible and they break the parsing of the sentence

2. Different types of brackets, you name them ( ), [ ], { }, of course it depends

3. Encoding, sometimes there are characters inside the text which should be set before, even one single comma (,) can stop a whole system

4. Some languages (e.g. Java, PHP) are confused with single quote sometimes. It is better to use regex like style for them

5. A mixture of above points is also possible, for example a foreign name inside an English text with an encoding and single quote inside it

6. My mentor does not agree that a really too long sentence could confuse the parser but I have seen it :) The question is though how long a really long sentence is which I cannot answer

So I suggest in order to start the text extraction, following could be beneficial:

1. Visual inspection of a sample of text to see if there are unwanted items like comments in brackets and the like

2. Setting encoding correctly or neutralizing the text, if possible, e.g. by converting them to plain text in Notepad, Gedit, Nano, etc. (We are not talking about big data)

3. Step #2 does not remove line feeds. If system needs to process a single sentence at a time, there may remained line feeds.

3. Replacing single quotes, etc with their regex

You should look at MIMIC:

Java native Regex engine is a horror for every Java developer. Why they include it in the release is a great puzzle.

There is a regex engine I used for the NLP processing from Apache: https://opennlp.apache.org/docs/.

My own favorite is though the regex engine in Python that simply works. However, you have to instantiate the Python engine first before you use regexps http://www.jython.org/jythonbook/en/1.0/JythonAndJavaIntegration.html

weka , rapid miner, sklearn python library are the easy to use for classification

Thank you sir for your valuable answers

The hard way is: convert PDFs to text (e.g. with Adobe IFilter),

then write a parser or rather a pattern matcher for possible

structures of papers. Unfortunately the font information

might get lost by conversion. Font info would help to identify

text segments. The grammar should be enriched with a

database of proper nouns. For example, if a section

contains some first names, the unknown words following

might be surnames. Numbers in the range 1900-2017 are

most likely publication years. The 'Keywords' keyword

follows the abstract. You will have to find lots of

similar heuristic rules.

Update 2. Oct. 2017: I'm currently experimenting

with a crawler in C# and have to read PDF as well.

With this example code it seems to be feasible.

Regards,

Joachim

Vicente: 

If you want to start with Information Extraction, I recommend you to take a look at "The Wiley's Handbook of Computational Linguistics and Natural Language Processing - Chapter 18: Information Extraction". 

Best, Farshad 

First You tag the sentence using any tagger.. Opennlp kit., etc. Then English sentence can be active or passive.. English sentence follows SVO structure. Identify Verb from the tag returns by the tagger.. Depending upon this before Verb Subject will be there after that object will be there.. You can search one of my paper based on types of sentence as semantic analysis 

I have used many text association techniques for understanding mathematical diagrams. 

There are many generic methods such as machine learning clustering using flagged image data set, pre-cluster text labels based on patterns. I am not aware of any successful generic text association work to understand diagrams.

But generic text association techniques are not enough to extract knowledge from diagrams such as mathematical diagrams (Euler Digrams, Coordinates Graphs) which are in generic image formats such as jpeg or SVG. For successful text association, optimized techniques have to used using domain knowledge of the diagram type.

I have successfully tested some methods with understanding several types of mathematical drawings. Text association techniques in diagrams are bound to the domain knowledge of the specific diagram type.

Hi Maryam,

It's based on the idea behind collecting the data. Rating different time means each time reflexes a particular instance feature. Generally, we don't recommend using the mean for such case, but instead of that, we recommend considering them as independent data features/variables that all of them are required.

HTH.

Samer 

 Each cluster validity method varies in its evaluation, some methods calculates the intra cluster similarity and some evaluates inter cluster distances or centroids. The purpose of using the right index method depends on our data. In case of Silhouette method, it’s always better to choose Silhouette coefficients nearer to 1, which says it is a good number to choose.

As well, not all the APIs take into account the same features to determine the polarity of text. For example, some APIs make uses of ponctuation or emoticons, while some other no. Maybe these article could lead you some clues about which system you should take:

Check link below for datasets from facebook, google+, twitter, and other social networks.

MG4J uses Bloom Filter: http://mg4j.di.unimi.it/

From the documentation: "Optionally, an index cluster may provide Bloom filters to reduce useless access to local indices that do not contain a term."

If I understand correctly for word extraction you can use simple TF-IDF. Also try latent semantic analysis (LSA) and latent Dirichlet allocation (LDA) for topic modeling. But if you're going to do (name entity recognition) NER then I think you need to develop and train your own model, probably based on hidden Markov.

Irfan,

Take a look at :

A Large Benchmark Dataset for Web Document Clustering by  Mark P. Sinka and David W. Corne

Some of the challenges in deriving a dataset for information retrieval is the dependence of the initial selection of documents based on search engines. The best thing to do is to extract samples from as many search engines as possible to avoid bias.

Regards

Hi

Maybe it is too late to tell you that, but also you can find datasets on this page if you like

Goodluck

The problem is, identification of persons with

mental illness requires common sense. For example,

if a person says: tomorrow I'll eat hundred whales

for breakfast. A computer program isn't, at least

not now, able to identify that as a joke or

sign of insanity.

Articles:

Regards,

Joachim

Phillipa:

In the mid 1980's we used a software product called "Mouselab", a computerized Information Display Board system run on micro-computers. Since then I imagine there are versions that can be used ONLINE over the internet.

The Use internet for Educational Research:

Since internet is a public domain and no one guaranty and responsible of what are being wrote or spread. It may be an obsolete knowledge or information, or it is partially tested information or it is uncompleted information. Therefore, it is higher risk to consume internet information as base for academicals writing. Especial the information whereas is spread over the anonymous sites.

Although there were anonymous and uncompleted sites, we can found millions of popular journals, economic magazines, and bulletins such as New York Post, Strait times, Washington post, universities libraries etc.

Reference

The Term Document matrix {X} has decomposed using SVD as follows:

{X}={W}{S}{P}'

After this these decomposed matrices have been reduced and the product of these reduced matrices has been calculated as {X`}. Then the similarity between two terms have been calculated using correlation between the vectors of terms from this matrix.

I have a question: Can we calculate term similarity only from {W} matrix after reduction? Is {W} matrix also give similar type of relation between terms as {X}?

Hi,

I'm not sure if software documentation qualifies for technical documentation, but Opus project has parallel corpora for PhP, Gnome, Kde and Ubuntu manuals:

Hope this helps.

Hello Vaishali you can check these URLs may be it can help you:

Good luck

Mussa,

This might not be a very helpful answer, but from my understanding NLP techniques often rely on context to understand what is being discussed.  So a single word like "doctor" is very difficult to understand unless it is in some kind of context like "a doctor treats sick people".  From the sentence, an NLP machine might recognize that doctor is a noun and might infer something about relating to people.  Without this context, it will be tough to discern the categorical differences between single words. 

It might be less complicated (although more time-consuming) to create a predefined list of terms that you would like to classify and then simply match words to those lists in order to create your associated list of features for a given entity.

Hope that helps.

Sean

In the European project MANTRA we developed a parallel corpus based on EMEA and MedLine. It comprises of a silver standard large corpus (using different systems tagging the corpus with a majority vote) and a manually crafted 550 document gold corpus (only EMEA) used to calculate precision and recall of the silver standard corpus. The link for the downloads is https://sites.google.com/site/mantraeu/project-output and project information can be found at https://sites.google.com/site/mantraeu/

I finally manager to sum up the system we have created for measuring the rotational speed of an object. Some initial assumptions:

1. The system must be mobile - in order to achieve this we combined our proprietary sensor with a smartphone.

2. The application is used for recording and evaluating received values of the rotational speed.

3. The application selects the maximum value form 2 second time window which is used for data aggregation.

Unfortunately we have not been able to produce proper results on built-in android sensors. We needed to develop a dedicated sensor solution which have than been integrated through Bluetooth with a smartphone.

 I will attach some dev specs to show how.

@ Marc Le Goc

Abstraction is a part of the blending process for sure. Especially during the construction of the Generic Space. I haven't come across the term "Knowledge Engineering" before. It sounds like a pretty interesting field. :)

@ Ignacio Arroyo

I haven't got into annotation schemes yet. But since CIT has a strong 'evolutionary' undercurrent running through it., they'll need to reflect it somehow.  A simple static semantic tag won't work. I'm thinking something more along the lines of vectors and graphs. 

information which is incomplete is not the same as missing data. In simple terms, "Incomplete information" means that we do not have enough *information* (attributes) to describe the underlying concept in order to be able to say which objects belong to it with certainty. An equivalence relation is used in order to build information 'granules' which can then be used to measure indiscernibility between data objects as a way to try to distinguish between them.  Missing data is just that: data that is not available or has been lost or corrupted in the dataset for whatever reason. RST can still be used to do data imputation based on the indiscernibility but this is a rather different task. 

sorry I do not have the expertise to answer this question.

Regards

Dear Lohweg,

The authors of the following paper:

that I tried to summarize and review:

choose to increase the model width instead of depth, because deeper models have only diminishing improvement or even degradation on accuracy. 

I believe that there might be some theoretical lower bound that determine the minimum number of layers given a raw features, whereas increasing the number of layers will only give minimal advantages or even degradation. But I don't know if such lower bound exists.

Hello Pawel,

Look into the attached paper. & this link for the application.

This paper gives you the well structured idea about application and extension of the Grid technology to knowledge discovery in Grid databases.

If you are working on larger Datasets, I'm certain OLAP could help you with it and provide better results than any other.

Furthermore, you also need to work on the performance results & usability of such applications.

Regards,

Manish

Hi, chandra Prakash, I mean I would like to analyze texture features of segmented objects, but it will generate much texture features for each band because of a lot of bands. I don't know if it is feasible to integrate several bands, and how to carry it out? 

I hope this will help. 

Good luck Riyadh ;)

Did you try the classics: cross-validation of your data set?

Hi, you can give weight distribution for each approaches. You can simply add all the scores of each approaches, then you can re-rank it again and get the best solution. So, first you need to specify the appropriate weight for each approach.

You can try Brand24 the most famous product which is the tool for monitoring the Internet and social media

Hi Nissi Paul,

Please go through following links, there you can find human detection source code.

Dear Sir,

Yes, that is correct.

The goal of OFDM is to enlarge the symbol duration T. In order not to lower the data rate, multiple carriers are then used. E.g. suppose symbol duration T, then set

w = 2PI/T, then the carriers used are

exp(jkwt), k = 0 ... N

You can see these carriers are orthogonal, meaning that if you multiply any 2 of these carriers, and integrate the result over T, then the result is always zero.

To transfer information, these carriers are modulated. Typically, QPSK, QAM16, QAM64, QAM256 modulation is used.

Note that not all carriers need to be modulated. Sometimes, pilots are sent on some to help in synchronization.

Note the main advantage of OFDM is immunity for multipath distortion. (dispersion.) Due to dispersion/multipath , there is a variable delay in the receive signal, there being a max time difference of delta-t between the first received signal, and the last. Now, in order avoid intersymbol interference, the transmit symbol duration is at least delta-t larger than T. In this case, it is possible in the receiver to recover the received symbols without intersymbol interference. There is still a need to compensate the frequency-dependent fading, but this is rather easy.

You easily see, to have efficient transmission in case of rather long delta-t, you need long T, and then in order to have sufficient data throughput, you need many carriers.

(rather large N).

Best Regards,

Henri.

Dear Noah,

Here are some papers with a short summary of their best feature sets for opinion mining + their references.

Pang et al. (2002) reported that unigrams outperform bigrams when performing the sentiment classification of movie reviews,

Dave et al. (2003) show that bigrams and trigrams worked better from unigrams for the product-review polarity classification.

Pak and Paroubek (2010) showed that their classifier is able to determine positive, negative and neutral sentiments of documents. Their classifier is based on the multinomial Na¨ıve Bayes classifier that uses N-gram and POS-tags as features.

Full references:

Kushal Dave, Steve Lawrence, and David M. Pennock.

2003. Mining the peanut gallery: opinion extraction

and semantic classification of product reviews. In WWW

’03: Proceedings of the 12th international conference on

World Wide Web, pages 519–528, New York, NY, USA.

ACM.

Bo Pang, Lillian Lee, and Shivakumar Vaithyanathan.

2002. Thumbs up? sentiment classification using machine

learning techniques. In Proceedings of the Conference

on Empirical Methods in Natural Language Processing

(EMNLP), pages 79–86.

Pak, A., & Paroubek, P. (2010, May). Twitter as a Corpus for Sentiment Analysis and Opinion Mining. In LREC (Vol. 10, pp. 1320-1326).‏

Best regards,

Yaakov

tutorial

Consider the following code

Binding of variables for use in compiling patterns:

Use Env env = TokenSequencePattern.getNewEnv() to create a new environment for binding

Bind string to attribute key (Class) lookup: env.bind("numtype", CoreAnnotations.NumericTypeAnnotation.class);

Bind patterns / strings for compiling patterns

// Bind string for later compilation using: compile("/it/ /was/ $RELDAY");

env.bind("$RELDAY", "/today|yesterday|tomorrow|tonight|tonite/");

// Bind pre-compiled patter for later compilation using: compile("/it/ /was/ $RELDAY");

env.bind("$RELDAY", TokenSequencePattern.compile(env, "/today|yesterday|tomorrow|tonight|tonite/"));

Your question is very confusing.  First you say that MaxEnt > SVM for NLP but then you say that SVM > MaxEnt.

Is there a clarification you could add to your question?

Also, if you are using linear classifiers, I would strongly recommend checking out L1 regularized logistic regression.

If you are using kernel methods to get non-linear classifiers, check out a recent implementation of neural networks with drop-out regulariztion, rectified linear units and accelerated learning like Adagrad.  You should get superior result versus your SVM or MaxEnt result.

Speaking as a Mahout committer, I would currently recommend several other options for k-means due to the passage of time,

1) check out H2O.  The guys at 0xdata have implemented a blazingly fast fork-join framework that allows very nice implementations of iterative algorithms like k-means.  They have a very nice implementation, in fact.

2) check out Spark.  Not nearly as advanced as H2O for k-means, but provides a very nice programming environment for parallel computing, Implementing k-means is relatively easy and Spark is now included in the major Hadoop distributions like MapR or Cloudera.

Thank you Fakhri again. Your advice is that we should divide the whole megalopolis into many subregion, then we research the rules in these subregion. Yes, that's a good idea! However, another question maybe we have to face with is that how to divede the megalopolis?  If we could not provide strong evidence about how to divide the megalopolis, our research results maybe questined. Someone could say that we get the results maybe just by accident. Do you have any idea about that?

Thanks so much!

The unsupervised classifications are tacitly built this way.

In case you intend to develop your own algorithm, try to select also windows of more pixels than one, for example 3x3 (it depends on the crown spread) and analyze the results as respect to the "one-pixel" training.

[ Reason: the spatial resolution of 1 m is too close to the target size, so the identification results may be affected if, for example, a major part of some training pixels is not covered by the supposed trees; the idea is to size a window for training so that it is mostly covered by the crown of the tree to be identified].

Hi Phuong, you can try many algorithm:

Many tools also you can try such as Weka, Matlab, RapidMiner, etc.

Being noted that instead of using only 2 class (positive and negative) in sentiment analysis. You can consider to employ the neutral label as the 3rd class.

This article and paper could be your further reference:

Good Luck

Hello Muhammad, 

I believe this quick tutorial should help you. The 2nd link I've included has further instructions on how to compile WordNet not included in the 1st link.

Co-occurrence of words with documents can also be used as a measure for semantic similarity. This was done, for example, in Explicit Semantic Analysis (ESA), http://www.cs.technion.ac.il/~gabr/resources/code/esa/esa.html

Dear Ines,

Though information extraction is highly data dependent, still Kernel based information extraction is more advanced approach for such cases. It certainly gives a better result compared to feature based information extraction.

(Unfortunately I could not attach the doc file containing the following text. Therefore I just pasted the text here.)

 First order histogram: (MATLAB)

      im = imread('tire.tif');

  nf = 20; featureVec = imhist(im, nf); % with any arbitrary value of nf

  % OR

  featureVec = imhist(im); % with default nf=256

 Second order histogram, here GLCM (MATLAB)

  im = imread('tire.tif');

  glcm = graycomatrix(im);

  temp = graycoprops(glcm);

  featureVec(1) = temp.Contrast;

  featureVec(2) = temp.Correlation;

  featureVec(3) = temp.Energy;

  featureVec(4) = temp.Homogeneity;

 First order histogram: (MATLAB)

  im = imread('tire.tif');

  [Nx, Ny] = size(im); % suppose the image is a single-band image

  w = 3; % neighborhood window --> (2w+1)-by-(2w+1)

  nf = 50;

  extIm = padarray(im, [w, w], 'symmetric');

  featureVec = zeros(Nx, Ny, nf); % memory allocation

  for x = 1+w:Nx+w

      for y = 1+w:Ny+w

         WIN = extIm(x-w:x+w, y-w:y+w);

         featureVec(x-w, y-w, :) = imhist(WIN, nf);

     end

  end

 Second order histogram, here GLCM (MATLAB)

   im = imread('tire.tif');

   [Nx, Ny] = size(im); % suppose the image is a single-band image

   w = 3; % neighborhood window --> (2w+1)-by-(2w+1)

   extIm = padarray(im, [w, w], 'symmetric');

   nf = 4; % graycoprops gives 4 features

   featureVec = zeros(Nx, Ny, nf);

   for x = 1+w:Nx+w

      for y = 1+w:Ny+w

         WIN = extIm(x-w:x+w, y-w:y+w);

         glcm = graycomatrix(WIN);

         temp = graycoprops(glcm);

         featureVec(x-w, y-w, 1) = temp.Contrast;

         featureVec(x-w, y-w, 2) = temp.Correlation;

         featureVec(x-w, y-w, 3) = temp.Energy;

         featureVec(x-w, y-w, 4) = temp.Homogeneity;

      end

   end

Hi Ayesha, I think you should first decide, which kind of industry you are interested in. One of the main challanges in process mining is getting access to the event log data. You can use publicly available logs as suggested by Marta. But it depends on what you want to do with that data and what kind of research question you intend to answer. Getting access to real event log data usually requires the cooperation of a partner from industry, and setting up this cooperation is usually hard work. I guess that you will probably not be able to get easy access to such data without a partner from practice. From my perspective searching for publicly available event log data is not a good approach to identify a new application scenario for process mining. I would start with the identification of a research problem maybe in a specific industry and then look for a partner from this industry that is willing to provide the necessary data.

Hi,

Another type of ontology is knowledge graph such as Freebase (https://www.freebase.com/), which allows users to download the weekly data dumps or use API to access the information.

best regards,

Generally, in the area of academic work, the core of programming languages are used. not software technologies.

but the decision about web-based(j2ee) or desktop-application based(j2se) should be made by through consideration of customer requirements. if you want to implement a package for classification and information extraction.

I suppose it is no longer relevant, but as you mentioned forum threads, I believe this work uses social networks of forum-like environments annotated for agreement/disagreement:

The data sets should be available at:

May be this paper could help you..

You can also forget the Stanford Tokens and try something like: "[Ee]ach .* ha(s|ve).*\." <-that's will return the first three sentences. 

Hi Ahmed,

In addition to imputation methods such as KNN, regression and MLP for missed data, other approaches can be applied. Ensemble classifiers such as AdaBoost and Bayesian Network can classify without imputation.

Please find the review paper as an attachment.

Two options:

or

These can be converted to R files quickly.

A simplified definition of a token in NLP is as follows: A token is a string of contiguous characters between two spaces, or between a space and punctuation marks. A token can also be an integer, real, or a number with a colon (time, for example: 2:00). All other symbols are tokens themselves except apostrophes and quotation marks in a word (with no space), which in many cases symbolize acronyms or citations. A token can present a single word or a group of words (in morphologically rich languages such as Hebrew) as the following token "ולאחי" (VeLeAhi) that includes 4 words "And to my brother".

A stirng as written by one of the previous researchers who responded

is a oncept taken from programming languages.

I think the previous two references provided by @Christos P Loizou and @Lucia Ballerini are new and excellent to start with.

No problem,may be it is used as benchmark to validate the develpped methods of for training.

I did not carefully read the whole paper, but maybe I can answer some of your questions.

How and how many classes should be eliminated?

I did not find explanations from the paper on this issue. A possible solution is to fix the number of classes to be selected, say between 50% and 75% of the whole classes. Then perform random class selection. The authors mentioned that "running PCA on a subset of classes instead on the whole set is done in a bid to avoid identical coefficients if the same feature subset is chosen for different classifiers." So basically, if you can ensure that you can avoid occasionally identical selection of features across the L classifiers, then you do not need to perform class selection, i.e. you can use the whole classes. Actually, even if you could not avoid identical feature selection, I personally do not think it is a big issue.

In every iteration a new random subset has to be selected?

The answer is yes, but as discussed above, I do not think it is a must.

What if it is a two-class data set?

Then obviously, you do not need to (cannot) perform class selection.

Does it make sense to select a bootstrap, centering the data to do PCA and then to generate scores using the rearranged rotation matrix on the whole data?

Yes, I think it makes sense. The idea is quite similar to random forest in the sense that both select a subset of features (feature dimensions if my understanding is correct) to introduce variation and thus avoid correlation between the L classifiers. The difference lies in the fact that rotation forest casts an additional PCA to the selected features. If we can consider feature selection itself as a simple feature projection process, then both random forest and rotation forest project the original feature onto a feature subspace, where rotation forest also translates (covariance-PCA), scales (correlation-PCA) and rotates the subspace. Rotation forest separates the features into subsets, and performs translation, scaling, and rotation to the subset features for both training and testing instances. It is similar to classification based on multiple types of features, where we perform PCA to each individual feature type, concatenate the projected feature vectors to obtain a single vector, and then use the single vector for training and testing. So yes, it makes sense to me.

1872 Orange pog.svg – Ludwig Boltzmann presents his H-theorem, and with it the formula Σpi log pi for the entropy of a single gas particle.

1878 Orange pog.svg – J. Willard Gibbs defines the Gibbs entropy: the probabilities in the entropy formula are now taken as probabilities of the state of the whole system.

1924 Red pog.svg – Harry Nyquist discusses quantifying "intelligence" and the speed at which it can be transmitted by a communication system.

1927 Orange pog.svg – John von Neumann defines the von Neumann entropy, extending the Gibbs entropy to quantum mechanics.

1928 Red pog.svg – Ralph Hartley introduces Hartley information as the logarithm of the number of possible messages, with information being communicated when the receiver can distinguish one sequence of symbols from any other (regardless of any associated meaning).

1929 Orange pog.svg – Leó Szilárd analyses Maxwell's Demon, showing how a Szilard engine can sometimes transform information into the extraction of useful work.

1940 Red pog.svg – Alan Turing introduces the deciban as a measure of information inferred about the German Enigma machine cypher settings by the Banburismus process.

1944 Red pog.svg – Claude Shannon's theory of information is substantially complete.

1947 Purple pog.svg – Richard W. Hamming invents Hamming codes for error detection and correction. For patent reasons, the result is not published until 1950.

1948 Red pog.svg – Claude E. Shannon publishes A Mathematical Theory of Communication

1949 Red pog.svg – Claude E. Shannon publishes Communication in the Presence of Noise – Nyquist–Shannon sampling theorem and Shannon–Hartley law

1949 Red pog.svg – Claude E. Shannon's Communication Theory of Secrecy Systems is declassified

1949 Green pog.svg – Robert M. Fano publishes Transmission of Information. M.I.T. Press, Cambridge, Mass. – Shannon–Fano coding

1949 Green pog.svg – Leon G. Kraft discovers Kraft's inequality, which shows the limits of prefix codes

1949 Purple pog.svg – Marcel J. E. Golay introduces Golay codes for forward error correction

1951 Red pog.svg – Solomon Kullback and Richard Leibler introduce the Kullback–Leibler divergence

1951 Green pog.svg – David A. Huffman invents Huffman encoding, a method of finding optimal prefix codes for lossless data compression

1953 Green pog.svg – August Albert Sardinas and George W. Patterson devise the Sardinas–Patterson algorithm, a procedure to decide whether a given variable-length code is uniquely decodable

1954 Purple pog.svg – Irving S. Reed and David E. Muller propose Reed–Muller codes

1955 Purple pog.svg – Peter Elias introduces convolutional codes

1957 Purple pog.svg – Eugene Prange first discusses cyclic codes

1959 Purple pog.svg – Alexis Hocquenghem, and independently the next year Raj Chandra Bose and Dwijendra Kumar Ray-Chaudhuri, discover BCH codes

1960 Purple pog.svg – Irving S. Reed and Gustave Solomon propose Reed–Solomon codes

1962 Purple pog.svg – Robert G. Gallager proposes Low-density parity-check codes; they are unused for 30 years due to technical limitations.

1965 Purple pog.svg – Dave Forney discusses concatenated codes.

1967 Purple pog.svg – Andrew Viterbi reveals the Viterbi algorithm, making decoding of convolutional codes practicable.

1968 Purple pog.svg – Elwyn Berlekamp invents the Berlekamp–Massey algorithm; its application to decoding BCH and Reed-Solomon codes is pointed out by James L. Massey the following year.

1968 Red pog.svg – Chris Wallace and David M. Boulton publish the first of many papers on Minimum Message Length (MML) statistical and inductive inference

1970 Purple pog.svg – Valerii Denisovich Goppa introduces Goppa codes

1972 Purple pog.svg – J. Justesen proposes Justesen codes, an improvement of Reed-Solomon codes

1973 Red pog.svg – David Slepian and Jack Wolf discover and prove the Slepian–Wolf coding limits for distributed source coding.[1]

1974 Red pog.svg – George H. Walther and Harold F. O'Neil, Jr., conduct first empirical study of satisfaction factors in the user-computer interface[2]

1976 Purple pog.svg – Gottfried Ungerboeck gives the first paper on trellis modulation; a more detailed exposition in 1982 leads to a raising of analogue modem POTS speeds from 9.6 kbit/s to 33.6 kbit/s.

1976 Green pog.svg – R. Pasco and Jorma J. Rissanen develop effective arithmetic coding techniques.

1977 Green pog.svg – Abraham Lempel and Jacob Ziv develop Lempel–Ziv compression (LZ77)

1989 Green pog.svg – Phil Katz publishes the .zip format including DEFLATE (LZ77 + Huffman coding); later to become the most widely used archive container and most widely used lossless compression algorithm

1993 Purple pog.svg – Claude Berrou, Alain Glavieux and Punya Thitimajshima introduce Turbo codes

1994 Green pog.svg – Michael Burrows and David Wheeler publish the Burrows–Wheeler transform, later to find use in bzip2

1995 Orange pog.svg – Benjamin Schumacher coins the term qubit and proves the quantum noiseless coding theorem

2001 Green pog.svg – Dr. Sam Kwong and Yu Fan Ho proposed Statistical Lempel Ziv

2008 Purple pog.svg – Erdal Arıkan introduced Polar Codes, the first practical construction of codes that achieves capacity for a wide array of channels.

Mussa,

ANSWER TO “How can I extract the adjective?”

--------------------------------------------------------------

You can extract adjectives from sentences by applying a parser, or more precisely, a part of speech tagger. Let me use your two sentences as example.

EXAMPLE 1. For the sentence:

“The large photo album has extra charges on delivery.”

1. You apply a parser in order to get the part of speech tags (nouns, verbs, adjectives, etc.) for each word. Suppose that we are applying the Stanford Parser to the above sentence. The parser will output the following:

The/DT large/JJ photo/NN album/NN has/VBZ extra/JJ charges/NNS on/IN delivery/NN

WHERE: JJ stands for adjective, NN stands for noun, VBZ stands for verb, and so on.

2. You extract all the words which have the JJ part of speech tag. For this sentence you extract:

- large/JJ

- extra/JJ

EXAMPLE 2: For the sentence/expression:

“Red car”

1. You apply the parser and you get the following output:

Red/JJ car/NN

2. You extract the words wich have the JJ part of speech tag:

- Red/JJ

PARSERS FOR JAVA:

---------------------------

In Java you can use the following free available parsers for the English language:

- The Stanford Parser: A statistical parser

- The OpenNLP library

(This library includes several tools for natural language processing. However you just need the “Part-of-Speech Tagger” for extracting adjectives).

- The TreeTagger

This tagger is not implemented in Java. Therefore if you want to use it from Java you need a wrapper. The tt4j is one of those Java wrappers for the TreeTagger: https://code.google.com/p/tt4j/

All the above links have all the information you need.

See:

Dear Hamed, you can do that using the WordNet (OWL version).

Use Protégé as a visualisation tool since it supports OWL ontologies

Thanks Hradis for your response with a reference.

My question is that each MSER-based candidate character can be like a window. Each window may have some features like height, width etc. What could be the other possible features that may help in detecting text?

Regards

You have a java syntax error here:

import edu.stanford.nlp.dcoref;

As 'dcoref' is a package, you should replace this line by:

import edu.stanford.nlp.dcoref.*;

To get the output a coreference resolution:

// Get coreference chains with:

Map<Integer, CorefChain> graph = annotation.get(CorefChainAnnotation.class);

// Then:

for(Map.Entry<Integer, CorefChain> entry : graph.entrySet()) {

System.out.println("ClusterId: " + entry.getKey());

System.out.println("CHAIN : " + entry.getValue());

CorefChain c = entry.getValue();

CorefMention representativeMention = c.getRepresentativeMention();

}