The benefit of the algorithm in the The idea of algorithms Algorithms NAME

project

Translated Sentence Mining

It is used to find sentences with the same describes the process of finding the
closest similar sentences between Translated
meaning from different languages. When we
search for a question in a specific language, we several different languages. For Sentence
Mining ‫اسراء رجب عبد‬1-
can find the answer in a source that is not from example, one set of sentences from one ‫الرازق عرفان‬
the language of the question. language and another set from a
different language. We want to find all
the similar sentences between the two
languages, so we used translated
sentence mining.
The overall project can benefit It is based on the idea image search ‫ أسماء عبدالباسط‬-2
and facilitate the search process, as it of converting the image to its ‫فتحى عبدالباسط‬
is possible to search for an image in vector and converting the text
this book with the help of this to its vector, then comparing
algorithm them and extracting the most
appropriate sentences for this
image
 Paraphrase mining is the task of
We use it to rephrase the question in more finding paraphrases (texts with
identical / similar meaning) in a ‫اشراق أشرف السيد‬3-
than one possible way in different terms, ‫عبد الحليم‬
large corpus of sentences
but with the same meaning Paraphrase
It is restatement of a text, passage,
It can also be used a lot in the translation Mining
or work giving the meaning in
process from one language to another
another form
because the translation gives the same
meaning in different ways, so we use it Given a list of sentences / texts,
when the question is in one language and
this function performs paraphrase
the answer is in another language, then it is
mining. It compares all sentences
translated into the question language to against all other sentences and
give the same desired meaning
returns a list with the pairs that
have the highest cosine similarity
score
 Cross- / Bi-
Bi-Encoder vs. Cross-Encoder:: Encoders

Bi-Encoders produce for a given

sentence a sentence embedding. ‫ فاطمة السعيد‬-4
We pass to a BERT independently ‫محمد شومان‬
the sentences A and B, which result
in the sentence embeddings u and
v. These sentence embedding can
then be compared using cosine
similarity: In contrast, for a Cross-
using a Cross-Encoder for semantic textual Encoder, we pass both sentences
similarity (STS). simultaneously to the Transformer
network. It produces than an output
We Known the similarity degree between
value between 0 and 1 indicating
the Question and answer.
the similarity of the input sentence
pair:

For example:

computes the score between a

query and all possible

sentences in a corpus using a

Cross-Encoder for semantic textual
similarity (STS).

It output then the most similar

sentences for the given query.
 Semantic search is a data searching
technique in a which a search ‫ أية كمال أيوب‬-5
query aims to not only find ‫محمد‬
used to improve search accuracy by Semantic
keywords, but to determine the
understanding the content of the search search
intent and contextual meaning of
query. the the words a person is using for
search.

It is provides more meaningful

search results by evaluating and
understanding the search phrase
and finding the most relevant
results in a website, database or
any other data repository.
 Retrieve & ‫ ايمان ايمن محمد‬-6
You can use this framework to compute ‫سليمان‬
sentence / text embeddings for more than 100
question answering retrieval improved Re-Rank
by using Retrieve & Re-Rank. we first
languages. These embeddings can then be
use a retrieval system that retrieves a
compared e.g. with cosine-similarity to find
large list of e.g. 100 we can use either
sentences with a similar meaning. This can be
lexical search, e.g. with ElasticSearch,
useful for semantic textual similar, semantic
or we can use dense retrieval with a bi-
search, or paraphrase mining.
encoder. A re-ranker based on a Cross-
Encoder
 ‫ ايمان محمد على‬-7
The cluster will be searshed for advanced Clustering
categories close to the answer and therefore the
speed of acess to the answer of the question. 1Convert data from mixed public data to
specific features

2Dividing mixed data into seprate

categories based on his youthful
qualities together.