File size: 13,888 Bytes
aa535a3 25d3bd1 1d6c0c4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 |
<!DOCTYPE html>
<html lang="en">
<head>
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Yelp LLM Query Understanding</title>
<script src="https://cdn.tailwindcss.com"></script>
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap" rel="stylesheet">
<style>
/* Use Inter font */
body {
font-family: 'Inter', sans-serif;
background-color: #f8fafc; /* Light gray background */
}
/* Custom styles for arrows and boxes */
.arrow {
position: relative;
width: 100%;
height: 2px;
background-color: #6b7280; /* Gray-500 */
margin: 1.5rem 0;
}
.arrow::after {
content: '';
position: absolute;
right: -1px;
top: -4px;
width: 0;
height: 0;
border-top: 5px solid transparent;
border-bottom: 5px solid transparent;
border-left: 8px solid #6b7280; /* Gray-500 */
}
.dashed-arrow {
position: relative;
width: 100%;
height: 2px;
background-image: linear-gradient(to right, #9ca3af 33%, rgba(255,255,255,0) 0%); /* Gray-400 dashed */
background-position: bottom;
background-size: 8px 2px;
background-repeat: repeat-x;
margin: 1rem 0;
}
.dashed-arrow::after {
content: '';
position: absolute;
right: -1px;
top: -4px;
width: 0;
height: 0;
border-top: 5px solid transparent;
border-bottom: 5px solid transparent;
border-left: 8px solid #9ca3af; /* Gray-400 */
}
.highlight-topic { background-color: #bfdbfe; color: #1e40af; } /* Blue */
.highlight-location { background-color: #bbf7d0; color: #15803d; } /* Green */
.highlight-time { background-color: #fef08a; color: #a16207; } /* Yellow */
.highlight-name { background-color: #fed7aa; color: #9a3412; } /* Orange */
.highlight-question { background-color: #fbcfe8; color: #9d174d; } /* Pink */
.highlight-spellcorrect { text-decoration: underline wavy #dc2626; } /* Red wavy underline */
.tag {
display: inline-block;
padding: 0.1rem 0.4rem;
border-radius: 0.25rem;
font-size: 0.75rem;
font-weight: 500;
margin-right: 0.25rem;
white-space: nowrap;
}
.llm-box {
border: 2px solid #fbbf24; /* Amber-400 */
background-color: #fefce8; /* Amber-50 */
}
.rag-box {
border: 2px dashed #a5b4fc; /* Indigo-300 */
background-color: #eef2ff; /* Indigo-50 */
}
.input-box, .output-box {
border: 2px solid #9ca3af; /* Gray-400 */
background-color: #ffffff; /* White */
}
.card {
background-color: white;
border-radius: 0.75rem; /* lg */
box-shadow: 0 4px 6px -1px rgb(0 0 0 / 0.1), 0 2px 4px -2px rgb(0 0 0 / 0.1);
padding: 1.5rem; /* p-6 */
margin-bottom: 2rem; /* mb-8 */
}
.section-title {
font-size: 1.5rem; /* text-2xl */
font-weight: 600; /* font-semibold */
margin-bottom: 1rem; /* mb-4 */
color: #1f2937; /* Gray-800 */
}
.step-label {
font-size: 0.875rem; /* text-sm */
font-weight: 500; /* font-medium */
color: #4b5563; /* Gray-600 */
margin-bottom: 0.5rem; /* mb-2 */
text-align: center;
}
</style>
</head>
<body class="p-4 md:p-8">
<h1 class="text-3xl md:text-4xl font-bold text-center mb-8 md:mb-12 text-gray-900">
Visualizing Yelp's LLM Query Understanding
</h1>
<div class="card">
<h2 class="section-title">1. Query Segmentation & Spell Correction</h2>
<p class="text-gray-700 mb-6">The LLM identifies and labels semantic parts of a search query, correcting misspellings simultaneously.</p>
<div class="grid grid-cols-1 md:grid-cols-5 gap-4 items-start">
<div class="flex flex-col items-center">
<div class="step-label">Input Query</div>
<div class="input-box p-3 rounded-lg w-full text-center shadow">
"healthy fod near me"
</div>
</div>
<div class="flex items-center justify-center md:mt-10">
<div class="arrow w-16 md:w-full"></div>
</div>
<div class="flex flex-col items-center">
<div class="step-label">LLM Processing</div>
<div class="llm-box p-3 rounded-lg w-full text-center shadow mb-2">
<span class="font-semibold text-amber-700">LLM</span>
<div class="text-xs text-amber-600">(e.g., GPT-4 / Fine-tuned Model)</div>
</div>
<div class="step-label mt-2">Retrieval Augmented Generation (RAG)</div>
<div class="rag-box p-2 rounded-lg w-full text-center shadow text-xs">
<span class="font-semibold text-indigo-700">RAG Input:</span>
<div class="text-indigo-600">Viewed Businesses: ["Salad Place", "Green Bowl"]</div>
</div>
</div>
<div class="flex items-center justify-center md:mt-10">
<div class="arrow w-16 md:w-full"></div>
</div>
<div class="flex flex-col items-center">
<div class="step-label">Output Segmentation</div>
<div class="output-box p-3 rounded-lg w-full text-center shadow">
<span class="tag highlight-topic">topic</span> <span class="highlight-spellcorrect">healthy food</span> <span class="tag highlight-location">location</span> near me
<div class="text-xs text-red-600 mt-1">[Spell Corrected: fod -> food]</div>
</div>
</div>
</div>
<div class="grid grid-cols-1 md:grid-cols-5 gap-4 items-start mt-12 pt-8 border-t border-gray-200">
<div class="flex flex-col items-center">
<div class="step-label">Input Query</div>
<div class="input-box p-3 rounded-lg w-full text-center shadow">
"pet-friendly sf restaurants open now"
</div>
</div>
<div class="flex items-center justify-center md:mt-10">
<div class="arrow w-16 md:w-full"></div>
</div>
<div class="flex flex-col items-center">
<div class="step-label">LLM Processing</div>
<div class="llm-box p-3 rounded-lg w-full text-center shadow mb-2">
<span class="font-semibold text-amber-700">LLM</span>
<div class="text-xs text-amber-600">(e.g., GPT-4 / Fine-tuned Model)</div>
</div>
<div class="h-12"></div>
</div>
<div class="flex items-center justify-center md:mt-10">
<div class="arrow w-16 md:w-full"></div>
</div>
<div class="flex flex-col items-center">
<div class="step-label">Output Segmentation</div>
<div class="output-box p-3 rounded-lg w-full text-center shadow">
<span class="tag highlight-topic">topic</span> pet-friendly <span class="tag highlight-location">location</span> sf <span class="tag highlight-topic">topic</span> restaurants <span class="tag highlight-time">time</span> open now
</div>
</div>
</div>
<p class="text-sm text-gray-600 mt-8">
<span class="font-semibold">Benefit:</span> This segmentation helps refine search location (e.g., understanding "sf" means San Francisco), improve business name matching, and apply relevant filters automatically. Spell correction ensures misspelled queries still yield good results. RAG helps distinguish business names from topics/locations.
</p>
</div>
<div class="card">
<h2 class="section-title">2. Review Highlights Generation</h2>
<p class="text-gray-700 mb-6">The LLM generates a list of related phrases for a query. These phrases are then used to find and highlight relevant snippets in user reviews, helping users quickly see why a business matches their search.</p>
<div class="grid grid-cols-1 md:grid-cols-5 gap-4 items-start">
<div class="flex flex-col items-center">
<div class="step-label">Input Query</div>
<div class="input-box p-3 rounded-lg w-full text-center shadow">
"dinner before a broadway show"
</div>
</div>
<div class="flex items-center justify-center md:mt-10">
<div class="arrow w-16 md:w-full"></div>
</div>
<div class="flex flex-col items-center">
<div class="step-label">LLM Processing</div>
<div class="llm-box p-3 rounded-lg w-full text-center shadow mb-2">
<span class="font-semibold text-amber-700">LLM</span>
<div class="text-xs text-amber-600">(e.g., GPT-4 / Fine-tuned Model)</div>
</div>
<div class="step-label mt-2">Retrieval Augmented Generation (RAG)</div>
<div class="rag-box p-2 rounded-lg w-full text-center shadow text-xs">
<span class="font-semibold text-indigo-700">RAG Input:</span>
<div class="text-indigo-600">Relevant Categories: [Theater District Restaurants, Pre-Theater Menus]</div>
</div>
</div>
<div class="flex items-center justify-center md:mt-10">
<div class="arrow w-16 md:w-full"></div>
</div>
<div class="flex flex-col items-center">
<div class="step-label">Output: Expanded Phrases</div>
<div class="output-box p-3 rounded-lg w-full text-left text-sm shadow">
<ul>
<li>- pre-show dinner</li>
<li>- theater district dining</li>
<li>- close to broadway</li>
<li>- quick bite before theater</li>
<li>- pre-theater menu</li>
<li>- convenient for shows</li>
</ul>
</div>
</div>
</div>
<div class="mt-12 pt-8 border-t border-gray-200">
<h3 class="text-lg font-semibold mb-3 text-gray-800 text-center">Application: Highlighting Review Snippets</h3>
<div class="flex flex-col md:flex-row items-center justify-center gap-4">
<div class="output-box p-3 rounded-lg shadow text-sm w-full md:w-1/3">
<span class="font-semibold">Generated Phrase:</span> "pre-show dinner"
</div>
<div class="flex items-center justify-center">
<div class="dashed-arrow w-16 md:w-10 transform md:rotate-0 rotate-90"></div>
</div>
<div class="bg-gray-100 p-4 rounded-lg shadow w-full md:w-2/3">
<p class="text-sm text-gray-700 italic">"... came here for a <strong class='bg-yellow-200 px-1 rounded'>pre-show dinner</strong> and it was perfect! Fast service and delicious food, just steps away from the theater..."</p>
<p class="text-xs text-gray-500 mt-1">- User Review Snippet</p>
</div>
</div>
</div>
<p class="text-sm text-gray-600 mt-8">
<span class="font-semibold">Benefit:</span> By generating semantically related phrases (not just keywords), the LLM helps surface highly relevant review snippets, even if the user's exact query words aren't present. This significantly improves the user's ability to quickly assess if a business meets their specific needs (like needing a quick dinner near a theater). RAG helps tailor phrases based on typical business categories associated with the query.
</p>
</div>
<div class="card bg-gradient-to-r from-blue-50 to-indigo-50">
<h2 class="section-title text-blue-900">Overall Yelp LLM Process (Simplified)</h2>
<div class="flex flex-col md:flex-row justify-between items-center gap-4 text-center">
<div class="p-3 bg-white rounded-lg shadow flex-1">
<div class="font-semibold text-blue-700 mb-1">1. Formulation & POC</div>
<p class="text-xs text-gray-600">Define task, prototype with powerful LLM (e.g., GPT-4), test on head queries.</p>
</div>
<div class="text-blue-600 font-bold text-2xl mx-2">→</div>
<div class="p-3 bg-white rounded-lg shadow flex-1">
<div class="font-semibold text-blue-700 mb-1">2. Scaling Up</div>
<p class="text-xs text-gray-600">Create golden dataset, fine-tune smaller/cheaper LLM (e.g., GPT4o-mini), pre-compute for high-volume queries.</p>
</div>
<div class="text-blue-600 font-bold text-2xl mx-2">→</div>
<div class="p-3 bg-white rounded-lg shadow flex-1">
<div class="font-semibold text-blue-700 mb-1">3. Production Rollout</div>
<p class="text-xs text-gray-600">Serve pre-computed results (cache), optionally use real-time smaller model (e.g., BERT/T5/GPT4o-mini) for tail queries.</p>
</div>
</div>
</div>
</body>
</html>
|