Cooking Reviews Segmentation and Classification based on

Deep Learning and Named Entity Detection

Randa Benkhelifa

and Nasria Bouhyaoui

2,1 b

Laboratoire de L'intelligence Artificielle et des Technologies de L’information, Université Kasdi Merbah,

Route de Ghardaia BP.511, 30 000, Ouargla, Algeria

Ecole Normale Supérieure de Ouargla, Ouargla, Algeria

Keywords: Text Segmentation, Text Classification, Online Social Network, YouTube, Cooking, Named Entity Detection,

Sentiment Analysis.

Abstract: YouTube is one of the most used online social networking (OSN) websites for exchanging recipes. It allows

uploading them, searching for, downloading, as well as rating and reviewing them. Sentiment analysis for

food and cooking recipes comments is to identify what people think about such cooking recipe video through

users’ comments. Nowadays, users’ give their opinion not only about recipes; they also evaluate the cook

through their comments, where a cook’s reputation can affect the users' opinion about his cooking recipes.

Frequently, when a cook has a good reputation, his recipes receive a great success by people, and vice versa.

In this paper, we propose a new approach that deal with the sentiment classification of cooking reviews.

Firstly, we examine the benefit of performing named entity detection and conjunctions on our corpus for text

segmentation in order to divide the comment on segments concerning the cook and segments concerning the

recipe. Next, we make two sentiment classifications (about the cook and about the recipe). Finally, we

incorporate the polarity of the cook sentiment classification in the recipe sentiment classification in order to

analyse the effect of the opinion about the cook on the performance of the categorization of the shared cooking

recipes comments in OSNs.

1 INTRODUCTION

Today, social networks such as Facebook, Twitter

and YouTube have become an essential element in

our daily life. Indeed, they are increasingly used to

convey messages and ideas through generating tons

of data on users and their interactions. The

importance of this data is that it contains a good

fraction of opinionated posts. Analyzing these posts

can reveal how users feel about certain topics, or

issues, events, products, people, recipes, etc.

Sentiment analysis is the field that allows the

treatment of users’ emotions and feeling. The

sentiment analysis is the process of determining

whether a piece of subjective writing is positive or

negative.

YouTube is a public video-sharing website where

people can share their experience and opinions in

order to maintain social relationships. One of the most

https://orcid.org/0000-0002-5577-4332

https://orcid.org/0000-0002-9810-3061

popular videos is the cooking videos, where, people

can upload, search for, download, as well as rate and

review recipes. Sentiment analysis of food recipe

comments is to identify what do people think about

such cooking recipe video through user’s comments

(positive or negative comments), where it is

interesting to predict their ratings automatically

(Benkhelifa and Laallam, 2018).

Frequently, the user does not give his/her opinion

only to evaluate the recipe; the user also gives his/her

opinion to evaluate the chef or the cook who prepared

this recipe. Sometimes, the judgment attributed to the

person who cook effects the opinion about the

cooking recipe.

Several papers over the years studied users’

opinions based on the textual content shared by them

in online social networks. The majority of the

previews works (Benkhelifa and Laallam, 2018). The

authors (Benkhelifa, et al., 2019) have focused on the

Benkhelifa, R. and Bouhyaoui, N.

Cooking Reviews Segmentation and Classiﬁcation based on Deep Learning and Named Entity Detection.

DOI: 10.5220/0011542300003318

In Proceedings of the 18th International Conference on Web Information Systems and Technologies (WEBIST 2022), pages 337-343

ISBN: 978-989-758-613-2; ISSN: 2184-3252

337

comments about the recipe ignoring user's opinion

about the cook, which can play an important role in

user opinion about the recipe.

In order to be able to process every part of opinion

separately; a part concerning the recipe and a part

concerning the cook, it is necessary to split the review

on segments. Text segmentation is a method of

splitting a document into smaller parts (Pak and Teh,

2018). This study is interested in only two types of

entities, which are person and recipe. Concerning

Person entity detection, lot of methods can be used to

annotate them. To the best of our knowledge, there is

no previous research concerning Recipe entities

detection. Therefore, we have created a list that

contains a great number of words concerning Recipe

NE such as (recipe, food, ingredient, vegetables,

cake, meat, meal, calorie, etc.).

In this paper, we propose a new approach, which

segment the review social text into two parts: the part

talking about recipe and the part talking about the

cook. We examine the benefit of performing named

entity recognition (NER) to an English corpus used

for text segmentation. Next, we build two sentiment

classifications, one for the recipe and the other for the

cook. After that, we study the effects of incorporating

the polarity of cook sentiment classification in recipes

sentiment classification. Finally, the experiments

show the impact of incorporating the opinion about

the cook on the performance of the categorization of

the shared cooking recipes comments in OSNs.

The remainder of this paper is as follows. In

section 2, the related works are discussed. In section

3, we present our methodology including our

proposed approach. The section 4 represents the

details of the results applied to the dataset extracted

from YouTube, and finally, we present a conclusion

and perspectives in section 5..

2 RELATED WORKS

Online social networking has become a part of the

daily routine of a huge number of persons (Benkhelifa

and Bouhyaoui, 2021). Comments in social media has

reached a great interest in several works on opinion

mining and sentiment analysis.

2.1 Text Segmentation

Text segmentation is a traditional NLP task, which is

widely used in text processing phase. It allows to

automatically partitioning text into coherent

segments or units. Each segment has its relevant

significance. Those units can be categorized as word

(Wu et al., 2007), (Liu and Chen, 2015), (Xia et al.,

2009) and (Zhang et al., 2021), sentence (El-Shayeb

et al., 2007), (Zhu et al., 2009), (Benkhelifa, et al.,

2019) and (Wicks and Post, , 2021), topic (Fragkou,

2013), (Ehsan and Shakey, 2016), (Memon et al.,

2021), (Lo et al., 2021), (Koshorek et al., 2018) and

(Maraj et al., 2021) or any information unit depending

on the text analysis task. The authors in (Memon et

al., 2021) have proposed a new topic-modelling-

based ensemble clustering approach, inducing the

combination of text segmentation and text clustering.

They have presented a cutting of a document into

segments (i.e. sub-documents), wherein each sub-

document is associated with exactly one sub-topic.

The work in (Lo et al., 2021), the authors have built

supervised neural text segmentation model in the

educational domain. A novel supervised training

procedure with a pre-labeled text corpus along with

an improved neural Deep Learning model for

improved predictions has been proposed in (Maraj et

al., 2021).

Recently, in text segmentation, many works such

as (Koshorek et al., 2018) and (Maraj et al., 2021)

have focused on supervised methods, which this is

formulated as a supervised learning problem

(Koshorek et al., 2018). The supervised methods have

solved two main drawbacks of the unsupervised

algorithms, which are the difficulty of specializing for

a given domain and in most cases, the unsupervised

methods do not naturally deal with multi-scale issues.

In addition, sentiment analysis has used text

segmentation in order to identify the polarity of each

segment. The authors in (Zhu et al., 2009) and

(Benkhelifa, et al., 2019) have used segmentation in

their model to identify multiple polarities and aspects

within one sentence. A novel method for aspect based

sentiment analysis, with an adaptation of LDA

Sentence Segmentation algorithm for product aspect

extraction has been proposed in (Ozyurt and Akcayol,

2021).

One of the essential and basic tasks of information

extraction and NLP is named entity recognition

(NER). Hence, the term ‘Named Entity’, is now

commonly used in NLP (Ramzi et al., 2017). NER is

the task of extracting, locating and classifying named

entities in a given piece of text. The named entity can

be a proper noun, a numerical expression which

represents type unit or monetary value, or a temporal

value which represents time. The classification of a

proper noun can be divided into three categories,

namely a person, a location, or an organization.

WEBIST 2022 - 18th International Conference on Web Information Systems and Technologies

338

2.2 Sentiment Classification

Recently, sentiment analysis is considered as the

process of finding users opinion about a particular

topic (Vijay et al., 2014). It performed on different

domain data such as Movie (Zhang and Zhu, 2006),

Books and Products (Dave et al., 2014), (Hu and Liu

, 2004), (Tan and Na, 2017) , Restaurants (Jingjing et

al., 2012), and cooking recipes (Liu, et al., 2014)

(Ning et al., 2013) (Benkhelifa, et al., 2019)

(Benkhelifa and Laallam, 2018), etc. The authors in

(Pugsee and Niyomvanich, 2015) have shown that the

reviews are the best rating predictors, in comparison

to ingredients, preparation steps, and metadata. Based

on users' reviews, various strategies for predicting

recipe ratings has been explored by the authors in

(Liu, et al., 2014). In order to improve the food

recipes a suggestion analysis method is proposed by

the authors in (Ning et al., 2013). A sentiment based

rating approach for food recipes using sorts food

recipes present on various websites from sentiments

of review writers is proposed in (Rao and Kakkar,

2017). The work in (Bianchini, et al., 2017) propose

a menu generation system is described, that takes into

account both user's preferences and healthy nutrition

habits. Another related research about food recipe

comments (Benkhelifa and Laallam, 2018) has

develop a real-time system to extract and classify the

YouTube cooking recipes reviews automatically.

This research has used the social media text

characteristics to improve the system performance.

The authors in (Benkhelifa, et al., 2019) has

introduced a sentiment based real-time system which

mines YouTube meta-data (Likes, Dislikes, views

and comments) in order to extract important cooking

recipes features and identify opinions polarity

according to these extracted features. Here we are

interested to show the effect of the cook’s reputation

on his cooking recipes reviews.

3 METHODOLOGY

The methodology followed in this work is presented

in this section. We start by introducing the description

of the process; how our method works.

3.1 Process

The process of the proposed approach is shown in

Figure 1.

Our method separates the comment on two parts

(cook’s review part and cooking review) using

text segmentation techniques. Then, we apply the

Figure 1: The process of the proposed approach.

sentiment classification on cook’s part comment. The

obtained polarity is combined to recipe review text

for recipe sentiment classification.

Comment: it represents a short social text

extracted from the well-known social network

YouTube, where any person can access and express

his/her opinion freely.

Comment segmentation: the segments of a

comment are divided into two bags, cook’s bag: this

bag represents the text review concerning the cook,

for example: “I love the way you cooked and smile

amazing chef!”. Where the second part represents

recipe bag, which reflect the user’s opinion about the

recipe itself, for example: “I want to eat this, Looking

delicious”.

After that, we look for the polarity of the cook’s

part comments; the value of this polarity is included

to the recipe sentiment classification process.

Here, we build the machine learning recipe

sentiment classification, the classifiers are not based

only on the textual content concerning the recipe they

consider also the value of the polarity of the cook’s

parts of comment, this value is token as input of the

classifier beside of the textual content.

3.2 Text Segmentation Approach

To build a robust system of comments segmentation,

we propose to use a supervised segmentation based on

named entity recognition. This study is interested in

only two types of entities, which are person PER

(representing the cook) and recipe REC. The other

entities (that does not describe a person, or a recipe)

including all other types of named entities NE such a

Cooking Reviews Segmentation and Classiﬁcation based on Deep Learning and Named Entity Detection

339

location, an organization, etc. are not taken into

account. Concerning Person entity detection, lot of

methods can be used to annotate them. To the best of

our knowledge, there is no previous research

concerning Recipe entities detection. Therefore, we

have created a list that contains a great number of

words concerning Recipe NE such as (recipe, food,

ingredient, vegetables, cake, meat, meal, calorie, etc.).

3.2.1 Text Segmentation Process

Our approach of segmentation is described by an

Algorithm, which is constituted of the following steps

as illustrated in Figure 2:

• Step 1: we use a separator segmentation: the

comment is split according to the punctuation.

• Step 2: for each segment, we detect the NE. If the

segment does not contain NE, we put it in the

recipe bag. Else, if the segment contains only one

type of NE, we put it in the bag of this entity type.

• Step 3: if the part of the comment contains more

than one kind of named entity, we split this

segment according to the conjunctions (except the

“and” conjunction, which is removed in the

preprocessing step). The result obtained could be

two segments or more. In order to categorize each

obtained segment into the adequate bag, we return

to the step 2.

Figure 2: The process of comment segmentation.

• As an output, we get two sets (bags) of segments

for each NE type (a cook’s bag and a recipe bag).

3.2.2 Text Segmentation Algorithm

Here, we give some notations. Suppose our dataset D

has

n Comments. Each comment c in C is segmented

to a set of part (unit) p in P. Each unit p has a label l

from L={l

per

rec

}. Where l

per

represents the unit

talking about the cook, and l

rec

is the label of the

recipe unit.

Algorithm comment_segmentation{

Inputs: a set of comments C, a set of

parts P, boolean b, a set of Named

Entities NE, a set of labels L;

Output: a cook bag Pper, a recipe bag

Prec;

For each comment c in C Do {

P= separator_segmantation(c);

}

For each part p in P Do{

b=Detect_entity(p);

if b==no then{

put p in Prec;

}

else{

NE=annotate (p);

if (same_label l(NE)==yes)

then{

if l(E)==REC then {

put p in Prec

}

else{

put p in Pper

}

else{

Conjunction segmentation;

Goto line 10;

}

Output (Prec, Pper);

4 EXPERIMENTS

4.1 Dataset Collection and

Construction

This section presents the dataset used in this work for

both segmentation and classification. The main

objective of this work is to segment and classify

cooking reviews. Moreover, to study the effect of

cook’s reputation on the users' opinion about his

cooking recipes and to show how much it is important

WEBIST 2022 - 18th International Conference on Web Information Systems and Technologies

340

to consider this information in the classification

process. YouTube comments are perfect for these due

to their abundance and a short length. Moreover,

YouTube is a popular video social network with a

great diversity of users, which means that collecting

a sufficiently large dataset with those characteristics

on various topics is feasible. To ensure the

consistency and the reliability of our proposed

approaches, we tested our classification and method

on a collection of 10000 recent texts of YouTube

comments about videos of cooking recipes collected

between (May and August 2016) from many

YouTube videos. Three human annotators as

following annotated these texts manually: for creating

the training model of the sentiment classification

(5000 positive and 5000 negative). Cooking recipe

corpus and its annotations guideline had been

originally defined in (Benkhelifa and Laallam, 2018).

4.2 Data Pre-processing

Preprocessing 1: it represents the prepressing, which

is applied before the segmentation algorithm.

1. Removing {and}, in the most cases the

conjunction “and” is used for listing several

description of the same entity for example

“she’s always happy and positive”.

2. Removing all words that can be annotated as

Person but do not represent the cook and

generally those words concern the comment

author such as {I, My kids, My family, etc.}

Preprocessing 2: it represents the prepressing, which

is applied before the classifications.

3. A term that appears less than three (3) times is

removed;

4. Removing punctuation (.,!?) and symbols

([<>());

5. The stemmer employed is the lovenStemmer,

which is used in the literature.

4.3 Evaluation Metrics

We adopt PK, the standard evaluation metric for text

segmentation for reporting the results of the prosed

text segmentation. PK score is the percentage of

wrong predictions on whether or not the first and last

sentence of a randomly sampled snippet of k

sentences belong to the same segment. Following the

previous works, we set k to the half of the average

ground truth segment size of the dataset.

Evaluation metrics for sentiment classifications

We now evaluate the performance of our method

by evaluating the performance of machine learning

training models, F-Measure (F) is one of the standard

metrics employed for evaluating our machine

learning models, this metric includes two

fundamental factors, i.e., precision (P) and recall (R),

which are obtained from the following relations:

P=TP/(TP+FP) (1)

R=TP/(TP+FN) (2)

F1=2*R*P/(R+P) (3)

5 RESULTS AND DISCUSSION

5.1 Text Segmentation Results

After the segmentation phase, we have gotten 12% of

PK.

We did not compare the obtained PK result with

the other researches, and this is because our main

objective is not the comment segmentation. In this

work, we are interested in the review sentiment

classification. The classification process required the

performing of text segmentation.

5.2 Sentiment Classification Results

Deep learning algorithms Convolutional Neural

Networks (CNN), and Recurrent Neural Networks

(LSTM, and bi-LSTM) are used for both sentiment

classifications. The first layer of each of these models

is a word embedding layer that turns sentences into a

feature map.

We start by showing the results of the cook’s

sentiment classification.

Table 1: Cook’s sentiment classification results.

Measure

Classifier

P R F

bi-LSTM

0.923 0.857 0.888

LSTM

0.888 0.843 0.865

CNN

0.86 0.86 0.86

The best F-measure result is 0.888. It has been

gotten using bi-LSTM classifier.

Now, we show the results obtained by cooking

recipes sentiment classification using three deep

learning algorithms CNN, LSTM, and bi-LSTM with

and without cook’s classification polarity.

TC: Textual Content.

OC: Opinion about cook.

Cooking Reviews Segmentation and Classiﬁcation based on Deep Learning and Named Entity Detection

341

Table 2: Recipe sentiment classification results.

Features Measure

Classifier TC OC P R F

CNN •

0.789 0.788 0.786

CNN • •

0.82 0.82 0.82

Bi-LSTM •

0.789 0.787 0.787

Bi-LSTM • •

0.83 0.826 0.825

LSTM •

0.82 0.811 0.81

LSTM • •

0.84 0.83 0.83

Comparison between the Gotten Results.

The highest results we got are 0.84 in precision, 0.83

in recall and 0.83 in F measure including both recipe

textual content and opinion about the cook using

LSTM classifier. Firstly, we have based only on the

textual content extracted comment concerning from

recipes without considering the opinion about the

cook, we got these classifiers precisions, recalls and

F-measures respectively, 0.789, 0.788, and 0.786

using CNN, 0.789, 0.787, 0.787 using bi-LSTM and

0.82, 0.81, and 0.81 using LSTM.

Including opinion about the cook (polarity value)

in the classification process, we remark an

improvement of 0.031 in precision, 0.032 in recall

and 0.034 in F-measure using CNN classifier, where

using bi-LSTM, we got an improvement of 0.041 in

precision, 0.039 in recall, and 0.04 in F-measure. The

LSTM classifier has gotten an improvement of 0.02,

0.019 and 0.02 in precision, recall and F-measure

respectively.

6 CONCLUSIONS AND FUTURE

WORK

In this paper, we have focused on text segmentation

and classification; we have proposed a new approach

of cooking comments segmentation based on deep

learning and NER. Where the segmentation has

gotten good results. In addition, we have shown the

impact of incorporating the opinions about the cook

on the performance of the classification of the recipes

comments extracted from YouTube. Thus, we

conclude that this integration has a good impact and

it plays a crucial role in improving the performance

of the recipes comments classification.

Future works: As a future extension of this work,

we plan to explore other characteristics for text

segmentation. We will also propose other approaches,

to improve the performance of OSNs text

classification.

ACKNOWLEDGEMENTS

This work was financially supported by "La Direction

Générale de la Recherche Scientifique et du

Développement Technologique (DGRSDT)" of

Algeria.

REFERENCES

Benkhelifa, R., & Laallam, F. Z. (2018, February). Opinion

extraction and classification of real-time youtube

cooking recipes comments. In International

Conference on Advanced Machine Learning

Technologies and Applications (pp. 395-404).

Springer, Cham.

Benkhelifa, R., & Bouhyaoui, N. (2021). Deep approach

based on user’s profile analysis for capturing user’s

interests. In International Conference on Artificial

Intelligence and its Applications (pp. 177-186).

Springer, Cham.

Benkhelifa, R., Bouhyaoui, N., & Laallam, F. Z. (2019). A

Real-Time Aspect-Based Sentiment Analysis System

of YouTube Cooking Recipes. In Machine Learning

Paradigms: Theory and Application (pp. 233-251).

Springer, Cham.

Bianchini, D., De Antonellis, V., De Franceschi, N., &

Melchiori, M. (2017). PREFer: A prescriptionbased

food recommender system. Computer Standards &

Interfaces, 54, 64-75.

Liu C, Guo C, Dakota D, Sridhar Rajagopalan,Wen Li,

Sandra K¨ubler, (2014). “My Curiosity was Satisfied,

but not in a GoodWay: Predicting User Ratings for

Online Recipes”, Proceedings of the Second Workshop

on Natural Language Processing for Social Media

(SocialNLP), pages 12–21, Dublin, Ireland.

Dave K, Lawrence S, Pennock D. (2003). “Mining the

peanut gallery: opinion extraction and semantic

classification of product reviews”. Proceedings of the

12th international conference on World Wide Web,

ACM, New York, NY, USA, WWW‟03.

Hu M, Liu B. (2004). "Mining and summarizing customer

reviews". Proceedings of the 10th ACM SIGKDD

international conference on knowledge discovery and

data mining. ACM, New York, NY, USA, KDD.

Jingjing Liu, Stephanie Seneff, and Victor Zue, 2012.

"Harvesting and Summarizing User-Generated Content

for Advanced Speech-Based HCI", IEEE Journal of

Selected Topics in Signal Processing, Vol. 6, No. 8,

pp.982-992.

Ning Yu, Desislava Zhekova, Can Liu, and Sandra K ubler.

(2013). Do good recipes need butter Predicting user

ratings of online recipes. In Proceedings of the IJCAI

Workshop on Cooking with Computers, Beijing, China.

Pugsee P., and Niyomvanich M., (2015) “Suggestion

Analysis for Food Recipe Improvement,” Proceeding of

the 2015 International Conference on Advanced

WEBIST 2022 - 18th International Conference on Web Information Systems and Technologies

342

Informatics: Concepts, Theory and Application

(ICAICTA).

Rao, S., & Kakkar, M. (2017). A rating approach based on

sentiment analysis. In Cloud Computing, Data Science

& Engineering-Confluence, 2017 7th International

Conference on (pp. 557-562). IEEE.

Tan, S. S., & Na, J. C. (2017). Mining Semantic Patterns

for Sentiment Analysis of Product Reviews. In

International Conference on Theory and Practice of

Digital Libraries (pp. 382-393). Springer, Cham.

Vijay B. Raut et al. (2014). “Survey on Opinion Mining and

Summarization of User Reviews on Web,” (IJCSIT)

International Journal of Computer Science and

Information Technologies, Vol.5 (2), pp.1026- 1030.

Zhang X., and Zhu F., (2006). “The influence of Online

consumer reviews on the demand for experience goods:

The case of video games”, In 27th international

conference on information systems (ICIS), Milwaukee,.

AISPress, 2006.

Pak, I., & Teh, P. L. (2018). Text segmentation techniques:

a critical review. Innovative Computing, Optimization

and Its Applications, 167-181.

Memon, M. Q., Lu, Y., Chen, P., Memon, A., Pathan, M.

S., & Zardari, Z. A. (2021). An ensemble clustering

approach for topic discovery using implicit text

segmentation. Journal of Information Science, 47(4),

431-457.

Maraj, A., Martin, M. V., & Makrehchi, M. (2021). A More

Effective Sentence-Wise Text Segmentation Approach

Using BERT. In International Conference on

Document Analysis and Recognition (pp. 236-250).

Springer, Cham.

Lo, K., Jin, Y., Tan, W., Liu, M., Du, L., & Buntine, W.

(2021). Transformer over Pre-trained Transformer for

Neural Text Segmentation with Enhanced Topic

Coherence. arXiv preprint arXiv:2110.07160.

Koshorek, O., Cohen, A., Mor, N., Rotman, M., & Berant,

J. (2018). Text segmentation as a supervised learning

task. arXiv preprint arXiv:1803.09337.

Zhang, D., Hu, Z., Li, S., Wu, H., Zhu, Q., & Zhou, G.

(2021). More than Text: Multi-modal Chinese Word

Segmentation. In Proceedings of the 59th Annual

Meeting of the Association for Computational

Linguistics and the 11th International Joint Conference

on Natural Language Processing (Volume 2: Short

Papers) (pp. 550-557).

Wicks, R., & Post, M. (2021). A unified approach to

sentence segmentation of punctuated text in many

languages. In Proceedings of the 59th Annual Meeting

of the Association for Computational Linguistics and

the 11th International Joint Conference on Natural

Language Processing (Volume 1: Long Papers) (pp.

3995-4007).

Wu, Y., Zhang, Y., Luo, S. M., & Wang, X. J. (2007).

Comprehensive information based semantic orientation

identification. IEEE NLP-KE 2007 - Proc (pp. 274–

279). Beijing: Int. Conf. Nat. Lang. Process. Knowl.

Eng. IEEE.

Liu, S. M., & Chen, J.-H. (2015). A multi-label

classification based approach for sentiment

classification. Expert Systems with Applications, 42,

1083–1093.

Xia, Z., Suzhen, W., Mingzhu, X., & Yixin, Y. (2009).

Chinese text sentiment classification based on granule

network. In: 2009 IEEE International Conference on

Granular Computing GRC 2009 (pp. 775−778).

Nanchang: IEEE.

Fragkou, P. (2013). Text segmentation for language

identification in Greek forums. In: Proceedings of

Adaptation of Language Resources and Tools for

Closely Related Languages and Language Variants (pp.

23–29). Hissar: Elsevier B.V.

Ehsan, N., & Shakery, A. (2016). Candidate document

retrieval for cross-lingual plagiarism detection using

two-level proximity information. Information

Processing and Management, 52, 1004–1017.

Zhu J, Zhu M, Wang H, Tsou BK (2009) Aspect-based

sentence segmentation for sentiment summarization.

In: Proceeding 1st International CIKM Worshop. Top

Analysis mass Open.— TSA’09 (pp. 65–72). Hong

El-Shayeb, M. A., El-Beltagy, S. R, & Rafea, A. (2007).

Comparative analysis of different text segmentation

algorithms on arabic news stories. In: IEEE

International Conference on Information Reuse and

Integration, Las Vegas (pp. 441–446).

Ramzi Salah, Lailatul Qadri Zakaria (2017) Arabic Rule-

Based Named Entity Recognition Systems Progress and

Challenges.

Ozyurt, B., & Akcayol, M. A. (2021). A new topic

modeling based approach for aspect extraction in aspect

based sentiment analysis: SS-LDA. Expert Systems

with Applications, 168, 114231.

Cooking Reviews Segmentation and Classiﬁcation based on Deep Learning and Named Entity Detection

343