Attention-based Ingredient Phrase Parser Zhengxiang Shi Pin Ni Meihui Wang To Eun Kim and Aldo Lipani University College London

2025-05-02 0 0 465.69KB 6 页 10玖币

侵权投诉

Attention-based Ingredient Phrase Parser

Zhengxiang Shi, Pin Ni, Meihui Wang, To Eun Kim and Aldo Lipani

University College London

Gower St, London - United Kingdom

Abstract. As virtual personal assistants have now penetrated the con-

sumer market, with products such as Siri and Alexa, the research com-

munity has produced several works on task-oriented dialogue tasks such as

hotel booking, restaurant booking, and movie recommendation. Assisting

users to cook is one of these tasks that are expected to be solved by in-

telligent assistants, where ingredients and their corresponding attributes,

such as name, unit, and quantity, should be provided to users precisely

and promptly. However, existing ingredient information scraped from the

cooking website is in the unstructured form with huge variation in the

lexical structure, for example, “1 garlic clove, crushed”, and “1 (8 ounce)

package cream cheese, softened”, making it diﬃcult to extract informa-

tion exactly. To provide an engaged and successful conversational service

to users for cooking tasks, we propose a new ingredient parsing model that

can parse an ingredient phrase of recipes into the structure form with its

corresponding attributes with over 0.93 F1-score. Experimental results

show that our model achieves state-of-the-art performance on AllRecipes

and Food.com datasets.

1 Introduction

There are few things so fundamental to our life as food, whose consumption is

intricately linked to our health, our feelings and our culture. With the rapid de-

velopment of science and technology, conversational AI has been a long-standing

area of exploration in the research community [1, 2, 3] and has now penetrated

in both academia and industries with products such as Microsoft Cortana and

Amazon Alexa. Recently, researchers work on integrating cooking tasks into

conversation systems with the target to assist customers to complete everyday

tasks [4]. To assist users with cooking tasks, ﬁne-grained information about each

recipe, such as cooking processes, utensils, nutritional proﬁle, dietary style, and

ingredient details, are needed. In particular, the ingredients details of a recipe

typically contain attributes such as quantity, temperature, and processing state.

Moreover, ingredient information itself can have use cases such as food pairing,

ﬂavour prediction, nutritional estimation, cost estimation and cuisine prediction.

In Figure 1, we show an example of a conversation where the intelligent agent is

assisting a user to cook. Here we can see that the quantity and other information

about the recipe are necessary to respond to users’ requests.

Although there are several datasets for the cooking domain, such as Recipe-

1M+ [5] and RecipeNLG [6], and a plethora of websites, such as AllRecipes

and WikiHow, providing plenty of human-readable recipes, there is a lack of

structured data about recipes useful to enable complex queries. Speciﬁcally,

arXiv:2210.02535v1 [cs.CL] 5 Oct 2022

How many ball mozzarella are needed to make Margherita Pizza?

About 125g.

User

Agent

Is there any requirement?

User

It should be sliced.

Agent

Thanks!

User Agent

Dialogue

Context

125g!ball

mozzarella

,!sliced

Database

Task-Oriented Conversational System Agent Process

Fig. 1: Applications of our ingredient parser model for conversational systems:

The user may ask for details of ingredients (left side). The agent will then retrieve

relevant recipes or cooking information from the pre-built database based on the

dialogue states and contexts (right side).

existing ingredient information and their corresponding attributes are in an un-

structured form with huge variations, for example, “1 garlic clove, crushed”, and

“1 (8 ounce) package cream cheese, softened”, making it diﬃcult to extract pre-

cise attribute information. As shown on the right side of Figure 1, the available

ingredient information that we can extract from existing databases is all in the

unstructured form. Addressing this problem needs the implementation of nat-

ural language processing algorithms that identify relevant attributes (quantity,

unit, temperature, processing state, etc.) from ingredient phrases. Diﬀerent

from common machine learning tasks, this ingredient phrase parsing task re-

quires high performance to provide users with more engaging and satisfactory

conversations. Diwan et al. [7] presented two dataset, AllRecipes and Food.com,

with 8 800 annotated ingredient phrases based on the RecipeDB [8]. However,

there is no ingredient parsing model with publicly available code on these two

datasets. To this end, we propose a novel attention-based neural network model

for ingredient parsing. Experimental results demonstrate that our model can

achieve state-of-the-art performance on AllRecipes and Food.com datasets.

2 Related Work

The neat deﬁnition of cooking tasks and the need to develop intelligent agents

has recently attracted much interest from the research community. Marin et

al. [5] proposed Recipe1M+ dataset, a large-scale, structured corpus of over one

million cooking recipes (including cooking instructions and ingredients) and 13

million food images. Based on this dataset, they trained a neural network to

learn a joint embedding of recipes and images for the image-to-recipe retrieval

task. Bien et al. [6] introduced RecipeNLG, a dataset of cooking recipes, built

upon Recipe1M+. RecipeDB [8] presents a structured, annotated dataset of

over 118,171 recipes, which are composed of 23,548 ingredients. The recipes

have been classiﬁed into cuisines represented by 26 geo-cultural regions that

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

Attention-basedIngredientPhraseParserZhengxiangShi,PinNi,MeihuiWang,ToEunKimandAldoLipaniUniversityCollegeLondonGowerSt,London-UnitedKingdomAbstract.Asvirtualpersonalassistantshavenowpenetratedthecon-sumermarket,withproductssuchasSiriandAlexa,theresearchcom-munityhasproducedseveralworksontask-orient...

展开>> 收起<<

Attention-based Ingredient Phrase Parser Zhengxiang Shi Pin Ni Meihui Wang To Eun Kim and Aldo Lipani University College London.pdf

共6页,预览2页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Attention-based Ingredient Phrase Parser Zhengxiang Shi Pin Ni Meihui Wang To Eun Kim and Aldo Lipani University College London

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: