In this paper, we introduce a recommender system where it automatically captures the context of what users or guests look for and recommends a bundle of products to be added to their shopping cart. The recommendation system takes selected recipes from a user as input and recommends a shopping cart with ingredients in optimized quantities as well as any kitchen gadgets that might be necessary to eficiently prepare the recipes using neural networks. We propose a system architecture, dive deep into the individual components, and evaluate the performance of information retrieval, semantic search, and quantity optimization algorithms. Using an ensemble methodology, we attained a mean average precision of over 0.9 for ingredient and quantity recommendations. The recipe-based bundle recommender system may be used not only to improve the user's shopping experiences but also to enable and encourage them to have healthier eating habits, aiming at providing personalized product recommendations.
Recommendation systems provide personalized recommendations for the customers using various data from customers and products to enhance customer experience and maximize the conversion rate, significantly contributing to revenue growth in the retail industry. In 2020, the recommendation system market was valued at 1.77 billion US dollars globally with a projected compound annual growth rate of 33.0% by 2028 [9].
Recommendation systems generally utilize two categories of algorithms, including Collaborative filteringmendation [16] algorithms. Collaborative filtering (CF) relies on the user’s history data and matches a user A with a similar user B and recommends A what B liked. Today, most big e-commerce giants with a massive amount of data, use collaborative filtering to recommend products to their customers [4]. Content-based recommendations create a profile to characterize each item. Some of the industry applications of content-based recommendation systems include a name that suggests jobs to the users by matching their interests and skills with the features of job postings [2]. IMDb uses the information of the movies present in this field in Section 2. Next, we provide the overview of the developed system, including mapping kitchen gadgets and ingredients in the recipes to relevant products from the product catalog database, optimizing the quantities to be recommended, and providing multiple candidates and their corresponding ranking to the user in Section 3. We have employed a few techniques to measure the performance of the system which will be discussed in Section 4. We summarize our work with some future works section 5. The authors make the codes used for the research available to the public [24]. customers can purchase individually or as a bundle, a 2. Related Work group of complementary items that can be purchased together [13] [29]. In late 2015, an American company that operates a gro
Our research ofers the convenience of personally cu- cery delivery and pick-up service in the United States rated shopping experiences to users by reducing their and Canada, integrated with AllRecipes, a top recipe site, shopping eforts to find out right products as a bundle. to allow users to select a recipe and fill their cart with all The system helps users without any prior experience the necessary ingredients [22]. Although the ingredient in cooking easily purchase the required ingredients and recommendations provided by a e-commerce company kitchen gadgets. An automatic quantity optimization are accurate for a good number of recipes, we observed will reduce wastage of resources and optimize costs for that the recommended quantities were not ideal for certhe users by suggesting the correct quantity of the prod- tain recipes. There were also cases where no matching uct. We believe that this approach leads to increased product was found for a recipe ingredient, such as veguser basket sizes, eventually raising the business revenue. etable oil, even though there are other closely related Furthermore, it can also aid in automated promotional products, including olive oil and canola oil. In addition emails to the consumers with all the ingredients and gad- to these limitations, we also identified an opportunity for gets in a recipe rather than the hand-curated contents augmenting ingredient recommendations with kitchen which we find time-consuming. gadget recommendations that could help improve a user’s
In this paper we discuss the related work already cooking experience, especially those new to cooking. Our work has been primarily motivated by these use cases, corpus (millions) while still being certain that recommenand in this paper, we propose a methodology that gen- dations are personalized and engaging for the user. Our erates accurate ingredient recommendations as well as research employed a two-stage approach for candidate kitchen gadget recommendations. generation (retriever stage) and ranking (re-ranker stage)
To our knowledge, there has not been any notable re- [7]. search regarding the recommendations of an optimized shopping cart of ingredients and kitchen gadgets based on recipes. We find that most of the existing literature in 3. System Overview the food domain is related to recommending recipes or ingredient substitutes. [25, 26, 11, 17]. Anirudh Jagithyala The proposed system first takes a recipe as input from [11] developed a recommendation system that recom- a guest shopping on an e-commerce company’s website. mends recipes based on recipe ratings, ingredients, and The recipe then gets split into two sections: 1) cooking review text. A number of approaches including memory- instructions and 2) ingredients. The cooking instructions based collaborative filtering and TF-IDF were tried along are parsed in order to detect and extract any kitchen with similarity measures such as cosine and Pearson cor- gadgets that might be required for the recipe, while the relation. The research evaluated multiple approaches us- required ingredients and quantities are extracted from ing the mean average precision (mAP) and showed that the recipe’s ingredients section. The quantities and units collaborative filtering on recipe ratings performed better. of ingredients in the recipe text and the product catalog Chantal Pellegrini et al. [17] explored the use of text and database are standardized for accurately matching varyimage embeddings for identifying ingredient substitutes. ing units from recipes and products. The ingredients and They generated context-free embeddings using word2vec kitchen gadgets required by the recipe are then fed into as well as context-based embeddings using transformer- the recommender system to search for the best matchbased models. In the end, the research showed that the ing products from the product catalog database based transformer-based multi-modal approach using text and on textual and quantity information. The system then image embeddings together gave the best results with a adds these products to the shopping cart of the guest precision of 0.84 for the top 1000 most common ingre- (Figure 2). dients. Chun-Yuen Teng et al. [25] explored the recom- For advanced natural language processing (NLP), we mendation of recipes and ingredient substitutes using utilized Open-source software libraries, including Spacy network structures. The system identifies ingredient sub- [10] and NLTK [3] for extracting recipe ingredients and stitutes using a graph structure where nodes represent in- kitchen gadgets from the recipe text. The ingredients, gredients and edges represent the degree of substitutabil- along with the required quantities, were parsed from the ity. To derive pairs of related recipes, they computed the ingredient section of the recipe (Figure 1) using regular cosine similarity between the ingredient lists for the two expressions, after which these ingredients were then prerecipes, weighted by the inverse document frequency. processed using the NLTK library for stop words removal, Mayumi Ueda et al. [26] applied user preferences and stemming, and -gram expansion. The Spacy library was ingredient quantity for recommending recipes. Their used for extracting kitchen gadgets from the recipe inmethod breaks recipes down into their ingredients and structions using named entity recognition (NER). Once scores them based on the ingredients’ frequency of use the ingredients and kitchen gadgets of the recipe are and specificity. identified, they are compared against the products in the
Some of the existing algorithms for searching and rank- product catalog database to get the most relevant proding relevant items use classical information retrieval al- uct in stock for each ingredient and user. This process gorithms such as TF-IDF [19], BM-25 [21], or Glove [18], also involved the novel usage of a language representawhile others make use of deep learning models such as tion model, Bidirectional Encoder Representations from BERT [8]. BERT [8] is a language representation model Transformers (BERT) [8], which is designed to pre-train that can give accurate contextual embeddings for words deep bidirectional representations from an unlabeled text in most cases. Unfortunately, the BERT does not gener- by jointly conditioning on both left and right contexts in ally give accurate representations for sentences, and the all layers. The advantage of using a pre-trained architecconstruction of BERT makes it unsuitable for semantic ture is that we can use transfer learning to transfer the similarity search. To overcome these issues, we applied already trained features to the current data without the the Sentence-BERT, [20] model, which was trained using complexity of training heavy machine learning models Siamese BERT-Networks. [15].
The preferred recommendation system architecture We also developed an algorithm to recommend the was one based a two-stage approach consisting of can- optimal number of products in case a guest chooses more didate generation and ranking stages. This two-stage than one recipe using the same products. The quantity approach allows for recommendations from a very large or weight of the common ingredients is recommended
The developed system utilizes a combination of semantic search and information retrieval algorithms to recommend the most relevant products for the ingredients in a recipe. Semantic search can improve search accuracy by understanding the context and content of the search which only find documents based on lexical matches, semantic search can also find synonyms. Semantic search aims to embed all entries in the corpus into vector space, where the query is embedded into the same vector space, to find the closest embeddings from the corpus. In our case, a recipe ingredient is a query, and the products are all the entries in the corpus, where the products are embedded into vector space and stored separately. During based on the sum of the required amount, which helps search time, the algorithm embeds a recipe ingredient create the optimal baskets for the guests and leads to less into the same vector space and calculates the similarity wastage of resources. query. In contrast to conventional search algorithms, , where is the embedding dimension of the ingredient between an ingredient ( ) and product ( ) to find the most relevant products. These products would have a high semantic overlap with the ingredient. We used cosine similarity in Equation 1 to find the closest embeddings from the corpus.
( , ) = ⋅ ‖ ‖‖ ‖ =
∑
=1 ⋅
√∑=1 2√∑=1 2
(1) and product vectors.
embedding of an ingredient and the embeddings of the products, the top products are retrieved.
For complex search tasks, the search can be significantly improved by using a retrieve and re-rank framework, where the top products are retrieved eficiently, followed by a re-ranker that ranks these products and 1, … , , the BM25 score for a product text is in Equa- levels: class, subclass, and item-type. These models were (2) function (Equation 5). , where ( ) is the number of product texts in the database that contain the term of the ingredient query. 3.1.2. Re-ranker (4) (5) recommends products. 3.1.1. Retriever Given an ingredient, we first use a retrieval system that quickly retrieves products that are potentially relevant for the given ingredient. Then the products are reranked, and the top
matches are sent through to the quantity recommendation module. For our retrieval system, we make use of the BM25 algorithm for lexical search [21]. For an ingredient query ( ) with terms tion 2.
25( , ) = ∑ ( =1 )
( , ) ⋅ ( + 1) ( ) + ⋅ (1 − + ⋅ || ) . , where ( , ) is the number of times term of the ingredient text occurs in . || is the number of words in
is the average number of words for product text. and are saturation parameters for document length and term frequency respectively. In general, values such as 0.5 < < 0.8 and 1.2 < < 2 are reasonably good in many circumstances [21].
Equation 3 describes inverse document frequency ( ( )) for a corpus with products with term .
( ) = − (
) + 0.5 ( ) + 0.5
combined the BM25 algorithm with a bi-encoder sentence transformer model that was fine-tuned using Microsoft’s
self-attention distillation to reduce the number of parameters required by a transformer model. It is twice as fast as BERT while retaining more than 99% accuracy on SQuAD 2.0 and several GLUE benchmark tasks using only 50% of BERT’s model parameters. The bi-encoder sentence transformer model [20] that uses the MiniLM model was trained using a dataset of 1 billion sentence pairs and a self-supervised contrastive learning objective: given a sentence from a sentence pair, the model should predict which out of a set of randomly sampled other sentences was actually paired with one in the dataset. A bi-encoder model performs two independent self-attentions for the query and the document, and the document is mapped to a fixed BERT representation regardless of the choice of a query. This makes it possible for bi-encoder models to pre-compute document representations ofline, significantly reducing the computational load per query at the time of inference [6]. A bi-encoder model can encode an input text, such as a recipe ingredient or a product, and output a vector (embedding) that captures semantic information. If a recipe ingredient embedding and a product embedding are similar, then the cosine similarity (Equation 1) between these two embeddings will be high.
the product embeddings using cosine similarity, we can identify the most similar products for an ingredient.
In order to reduce the search space eficiently, hierarchical classification models were created for the following trained using the preprocessed text of the products as feature vectors and the respective hierarchical level values as the target labels. A softmax activation function (Equation 4) was used in the final layer of the multi-class classification models, along with the cross-entropy loss Softmax ( ) = exp( ) ∑=1 exp( ) , where is the number of output classes, is an element of a vector of size corresponding to a particular class.
Cross Entropy Loss = −
(∑ ⋅ ( ̂ )) 1 =1 (3) , where is the number of observations, is the true label vector and ̂ is the predicted label probability vector. After retrieving the top products, the re-ranker stage ranks the products more accurately using a cross-encoder sentence transformer model that was fine tuned using the Microsoft’s MS Marco dataset [ 1], which is a large scale information retrieval corpus that was created based on real user search queries using Bing search engine. In contrast to a bi-encoder model that performs two independent self-attentions for the query and the document, a cross-encoder model performs full self-attention across the entire query-document pair. As a result, the crossencoder can model the interaction between a query and a document, and the resulting representations contain contextualized embeddings [6]. In our use case, an ingredient and a product are passed simultaneously to the cross-encoder, which then outputs a score indicating how relevant the product is for the given ingredient. As the cross-encoder models the interaction between an ingredient and a product during inference time, it is slower than the bi-encoder, and hence, it can only be used for a small subset of products. However, we can achieve a higher accuracy as they perform attention across the query and the document. After the re-ranker stage, the
of 1 lb rather than two packs of 0.5 lb flour if 1 lb-pack is
If a user selects multiple recipes, the quantity is optimized such that minimum units of the common products are recommended. For example, for two recipes using one tablespoon and two tablespoons of salt each, the recommended unit of salt can/bottle will be optimized to one to reduce waste and cost.
lar approach to ingredient recommendations using a combination of semantic search and information retrieval algorithms. The required kitchen gadget is implicitly menOnce the algorithm selects the top matched ingredients, tioned in unstructured recipe instruction text, whereas the next important step is to recommend the optimal ingredients are explicitly listed in the ingredient secquantity of the product needed in the recipe (Figure 4). tion. To identify the kitchen tools and methods post
specified in the recipe and normalize the units required to the pre-processing of the recipe instructions, a custom
a standard SI units (Table 2), including tablespoon (tbsp), The NER model identifies the kitchen gadgets (nouns) teaspoon (tsp), milliliter (ml), cup, count, pound (lb), and ounce (oz). These standard quantities are either weight or in volumes handled diferently from each other.
As product descriptions generally utilize weight as a measure while most recipes use volumes, we converted the volume with the standardized unit to weight using density ( ). For instance, the weight ( ) for cups of a grocery product in the recipe, where 1 cup is 225 ml, can be calculated as the following.
= ∗ 225 ∗
Once the required weight is calculated, the system compares it against the weight of the recommended products in product catalog’s database. The recommended number of units is then calculated for each matched product using Equation 7, where is the recommended quantity, is the ounces required in the recipe, and is the ounces sold or packaged.
= ⌈
⌉
a recipe, like two onions or three potatoes, the average weight of the given fruits and vegetables is used to convert it to weight [14]. The reverse conversion is also applied if the unit for a product at e-commerce companies uses count and the recipe specifies weight instead.
Once the recommended quantity is known for matched ingredients, we sort ingredients by the quantity recommended and the price in ascending order. In addition, the system recommends lower-priced items if multiple packaging options are available for the same products. For instance, the system recommends one pack (6) (7) used in the recipe and the methods (verbs) that can identify a kitchen gadget. For example, for ”Chop the garlic and add to the pan”, a pan will be identified as a gadget, whereas chop will be identified as a method associated with the gadgets including a knife and chopping board.
Similar to ingredient recommendations, products are embedded into vector space and stored separately. During search time, a gadget from the recipe instruction is also embedded into the same vector space, and the system searches for the most relevant products. These products would have a high semantic overlap with kitchen gadgets. After computing the cosine similarity between the embedding of the kitchen gadget and the embeddings of all products, the top products with the maximum similarity are retrieved. For these search tasks, we used embeddings from the RoBERTa [12], an improved and robustly trained version of BERT with further tuned hyperparameters. RoBERTa has achieved state-of-the-art results on GLUE, RACE, and SQuAD.
For complex search tasks, the search is significantly improved by using a retrieve (Section 3.1.1 ) and re-rank (Section 3.1.1) framework, just like for the ingredients (Figure 5). For quantity optimization, if the user selects multiple recipes, the common kitchen gadgets are only recommended once, along with all the other gadgets used in each recipe.
For ingredient search, diferent algorithms, as well as an ensemble of these algorithms, were evaluated using the
metric. The BM25 algorithm was considered as the baseline model against more complex models. An (9) algorithm gives a very good performance for =1 since lexical search algorithms generally have high precision due to exact keyword matching. However, for higher values of , the drop in mean average precision is quite high.
such as MiniLM and MS Marco are more general with consistently high precision values across diferent values. Using BM25 along with the MS Marco transformer model gave the best performance for all values.
, where is the number of queries, is the number of interesting thing to note from Figure 6. is that the BM25 at .
is the average precision ∑ =1
() =
∑ =1 ()
( , )
1
∑ () ⋅ () , where is the number of relevant products, is the number of retrieved products, () says whether that ℎ item was relevant ( () ( () =0), and () is the precision at .
is an indicator that
=1) or not
on 500 recipes and tested on 100 recipes. The custom entities’ kitchen gadgets and methods were marked with their position in the text using regex for these 600 recipes.
A manual review of the annotations was performed to update any incorrect or missed annotations, and we achieved a test F1 score of 99.627% (Equation 11). corpus of over one million cooking recipes and 13 million food images. The @1 value for all the ingredients from these 100 recipes was 0.949, which is similar to the @ values we see in Figure 6 for the top 100 ingredients.
2 ⋅ ⋅ 1 = (11)
+ , where precision is out of the total documents retrieved, how many are relevant and recall is out of the total rele4.2. Quantity Normalization and vant documents how many relevant documents are re
A custom mapping is used to convert cooking methods For evaluating quantity normalization, we measured the to kitchen gadgets which along with other identified gadaccuracy of quantity normalization, using the most com- gets form search queries for the recipe. For evaluation, monly used ingredients in the randomly chosen 100,000 diferent algorithms were evaluated at mAP@K, similar recipes. The three most commonly used ingredients, salt, to ingredient search. We developed transformer models sugar, and butter, are tracked to measure if the recom- including MiniLM, MS Marco, and Roberta for the semended quantity after unit normalization is accurate or mantic search tasks. The experiment results show that not. MS Marco, a combination of retriever and a re-ranker,
We found that salt is used in 32,506 out of the chosen performs consistently better than MiniLM and Roberta 100,000 recipes ( Table 3). Out of the total 32,506 recipes, across all (Figure 7). the quantity of salt is correctly recommended in 32,443 As an example, we present a recipe in Figure 8 that recipes which is 99.806% of the total. Similarly, the accu- consists of two sections: ingredients and directions. The racies of the recommended quantity were 82.435% and ingredients section is used for ingredient and quantity 78.463% for sugar and butter respectively. However, the recommendations, whereas the directions section is used relatively low accuracy in quantity recommendations for for kitchen gadget recommendations. The keywords used sugar and butter is due to the incorrect recipe text. For for kitchen gadget recommendations are highlighted in example, certain recipes say 34 cups of butter instead of red. 3/4 cups of butter. This has been further discussed in The product recommendations based on the recipe Section 5. (Figure 8) is given below in Table 4.
Further, the quantity optimization process was also evaluated with randomly chosen 100 recipes from the Recipe1M+ corpus. The mAP@1 value for all the ingredi- 5. Conclusion ents from these 100 recipes was 0.914. For a combination of recipes, the result has been manually verified for 5 random sets of any two recipes.
Table 4 assessing the top product recommendations for the 100 Shopping Cart Recommendation most frequently occurring ingredients in the 1M+ recipe Product Qty Price ($) corpus. For ingredient search, the ensemble approach of Red Delicious Apple 1 0.99 BM25 and MS Marco gave the best performance, while GMocoCdor&mGicakthGerroOunrgdaCniincnOaamtson-1-82o.3z7oz 11 21..3999 for kitchen gadget search, the MiniLM model proved to Buttermilk Pancake Mix - 32oz 1 2.19 be more accurate. The accuracy of quantity recommendaImperial Granulated Pure Sugar - 4lb 1 2.19 tions was measured by evaluating certain high-frequency SGqouoidsh&1G.5qatthMerixAinlkgaBlinoewWl-aGterree-n1L 11 08..9999 ingredients such as salt, sugar, and butter across more Anchor 8oz Glass Measuring Cup 1 3.49 than 50,000 recipes from the 1 million+ (1M+) recipe Nylon Ladle with Soft Grip - Made By Design 1 3.00 corpus.
FLaaukxesGidrean1i0t”e Nfinonisshtick Aluminum Skillet with 1 16.47 There were a few challenges that we faced while implementing the system that we hope to address in the near future. Firstly, the recipe text used in the 1M+ recipe by the recipe, which are extracted from the recipe text corpus has some inconsistencies where the quantity realong with associated quantities. More significantly, the quired is not accurately defined. For example, a lot of system is also capable of recommending the relevant recipes say 34 cups of an ingredient instead of 3/4 cups. In kitchen gadgets that might be required for making the future work, we can work on parsing the text with more recipe based on the instructions provided in the recipe. scrutiny and rules to avoid such cases. Alternatively, When a user selects multiple recipes, the recommender we could also build an entirely new recipe scraping alsystem optimizes quantities for each product, where the gorithm to handle such cases. Secondly, the quantity quantities are adjusted according to the amounts of the optimization for multiple recipes is in place, but there is common ingredients and kitchen gadgets present in the no formal way to measure how well is the performance recipes. of the process. As a part of future work, we would devise
We conducted experiments for evaluating the efec- a way to quantify the performance of this process. tiveness of various algorithms for the recommendation We also identified extra features for the current syssystem, such as BM25, MiniLM, MS Marco, and RoBERTa. tem. First, the dietary restrictions of a user can be acIn our experiments, we compared these algorithms by counted for by considering the ingredients, nutritional information, and key allergens that might be present in [5] Lin Chen, Rui Li, Yige Liu, Ruixuan Zhang, and the product using natural language processing. Second, Diane Myung-kyung Woodbridge. 2017. Mathere could be preference filters provided to users before chine learning-based product recommendation usthey add a recipe to the shopping cart that could consider ing Apache Spark. In 2017 IEEE SmartWorld, Ubiqtheir dietary preferences such as whether they prefer uitous Intelligence Computing, Advanced Trusted non-GMO or organic products only. Third, instead of Computed, Scalable Computing Communications, asking for preferences explicitly, the preferences of the Cloud Big Data Computing, Internet of People users could be determined automatically by analyzing and Smart City Innovation (SmartWorld/SCALtheir past shopping behaviors such as clicks, views, or COM/UIC/ATC/CBDCom/IOP/SCI). 1–6. ht tps : product purchases. Based on this implicit feedback, we //doi.org/10.1109/UIC-ATC.2017.8397470 could determine the user’s preferences, such as whether [6] Jaekeol Choi, Euna Jung, Jangwon Suh, and Wonthe user is a vegetarian or if the user is currently pur- jong Rhee. 2021. Improving Bi-encoder Docuchasing products specific to a particular diet, such as the ment Ranking Models with Two Rankers and Multiketogenic diet. Such information can then be used in the teacher Distillation. In Proceedings of the 44th Inrecommendation system for personalizing recommended ternational ACM SIGIR Conference on Research and products for diferent users. Development in Information Retrieval. ACM.
The content-based recommendation system proposed [7] Paul Covington, Jay Adams, and Emre Sargin. 2016. in this paper could potentially be used to improve the Deep neural networks for youtube recommendashopping experience of users by providing them options tions. In Proceedings of the 10th ACM conference on to add all the necessary products required by a recipe recommender systems. 191–198. automatically to their shopping cart. Kitchen gadget rec- [8] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and ommendations could be helpful not just for improving Kristina Toutanova. 2018. BERT: Pre-training of the shopping experiences of the users but also for their Deep Bidirectional Transformers for Language Uncooking experiences. These recommendations also al- derstanding. low a business to increase the basket sizes of their users, [9] Grand View Research. 2021. Recommendation Enthereby increasing revenue. Moreover, the system could gine Market Size, Share & Trends Analysis. Recomalso aid in automated promotional emails that recom- mendation Engine Market Report (2021). mend products required by recipes that may be of interest [10] Matthew Honnibal and Ines Montani. 2017. spaCy to customers. 2: Natural language understanding with Bloom embeddings, convolutional neural networks and incremental parsing. (2017). To appear.
References [11] Anirudh Jagithyala. 2014. Recommending recipes [1] Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, tbaatsieodn.onKainngsraesdSietnattes aUnndivuesresritryev.iews. Ph. D. DisserJianfeng Gao, Xiaodong Liu, Rangan Majumder, An- [12] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, drew McNamara, Bhaskar Mitra, Tri Nguyen, Mir Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Rosenberg, Xia Song, Alina Stoica, Saurabh Tiwary, Luke Zettlemoyer, and Veselin Stoyanov. 2019. and Tong Wang. 2016. MS MARCO: A Human Gen- RoBERTa: A Robustly Optimized BERT Pretraining erated MAchine Reading COmprehension Dataset. Approach. [2] Shivam Bansal, Aman Srivastava, and Anuja Arora. [13] Kevin F. McCardle, Kumar Rajaram, and Christo2017. Topic Modeling Driven Content Based Jobs pher S. Tang. 2007. Bundling retail products: ModRecommendation Engine for Recruitment Industry. els and analysis. European Journal of Operational Procedia Computer Science 122 (2017), 865–872. 5th Research 177, 2 (2007), 1197–1217. International Conference on Information Technol- [14] Niklas. 2022. Average Weight of All Fruits and ogy and Quantitative Management, ITQM 2017. Vegetables. https://weightofstuff.com/average-w [3] Steven Bird, Ewan Klein, and Edward Loper. 2009. eight-of-all-fruits-and-vegetables/ Natural language processing with Python: analyzing [15] Sinno Jialin Pan and Qiang Yang. 2010. A Survey on text with the natural language toolkit. ” O’Reilly Transfer Learning. IEEE Transactions on Knowledge Media, Inc.”. and Data Engineering 22, 10 (2010), 1345–1359. [4] Dan Chang, Hao Gui, Rui Fan, Ze Fan, and Ji [16] Michael J. Pazzani and Daniel Billsus. 2007. ContentTian. 2019. Application of Improved Collaborative Filtering in the Recommendation of E-commerce BHaesieddelRbeercgo,mBmerelnind,aHtioenideSlybsetregm, s3.25S–p3r4in1.ger Berlin Commodities. INTERNATIONAL JOURNAL OF [17] Chantal Pellegrini, Ege Özsoy, Monika Wintergerst, COMPUTERS COMMUNICATIONS CONTROL 14, 4 and Georg Groh. 2021. Exploiting Food Embeddings (2019), 489–502. for Ingredient Substitution. In HEALTHINF. [18] Jefrey Pennington, Richard Socher, and Christo- _project/ pher D Manning. 2014. Glove: Global vectors for [25] Chun-Yuen Teng, Yu-Ru Lin, and Lada A Adamic. word representation. In Proceedings of the 2014 con- 2012. Recipe recommendation using ingredient ference on empirical methods in natural language networks. In Proceedings of the 4th annual ACM processing (EMNLP). 1532–1543. web science conference. 298–307. [19] Juan Ramos et al. 2003. Using tf-idf to determine [26] Mayumi Ueda, Syungo Asanuma, Yusuke word relevance in document queries. In Proceed- Miyawaki, and Shinsuke Nakajima. 2014. Recipe ings of the first instructional conference on machine recommendation method by considering the users learning, Vol. 242. Citeseer, 29–48. preference and ingredient quantity of target recipe. [20] Nils Reimers and Iryna Gurevych. 2019. Sentence- In Proceedings of the international multiconference BERT: Sentence Embeddings using Siamese BERT- of engineers and computer scientists, Vol. 1. 12–14.
Networks. [27] Wenhui Wang, Furu Wei, Li Dong, Hangbo Bao, [21] Stephen Robertson and Hugo Zaragoza. 2009. The Nan Yang, and Ming Zhou. 2020. MiniLM: Deep Selfprobabilistic relevance framework: BM25 and beyond. Attention Distillation for Task-Agnostic CompresNow Publishers Inc. sion of Pre-Trained Transformers. arXiv:2002.10957 [22] Perez Sarah. 2015. Instacart And Allrecipes Now Let [28] Chutian Wei, Xinyu Chen, Zhenning Tang, and You Add A Meal’s Ingredients To Your Grocery List Wen Cheng. 2021. Fully content-based IMDb movie With A Click. https://techcrunch.com/2015/10/12/ recommendation engine with Pearson similarity. instacart-and-allrecipes-now-let-you-add-a-meals In International Conference on Green Communica-ingredients-to-your-grocery-list-with-a-click/ tion, Network, and Internet of Things (GCNIoT 2021), [23] J. Ben Schafer, Dan Frankowski, Jon Herlocker, and Siting Chen and Jun Mou (Eds.), Vol. 12085. InternaShilad Sen. 2007. Collaborative Filtering Recom- tional Society for Optics and Photonics, SPIE, 132 – mender Systems. Springer Berlin Heidelberg, Berlin, 137.
Heidelberg, 291–324. [29] Ruiliang Yan, Chris Myers, John Wang, and Sanjoy [24] Chahak Sethi, Melvin Vellera, Diane Myung kyung Ghose. 2014. Bundling products to success: The Woodbridge, and Joey Jonghoon Ahnn. 2022. Bun- influence of complementarity and advertising. Jourdle Recommender from Recipes to Shopping Cart. nal of Retailing and Consumer Services 21, 1 (2014), https://github.com/dianewoodbridge/target_recipe 48–53.